Video recommendation method, device, electronic device and storage medium

Abstract

The disclosed embodiment provides a video recommendation method, device, electronic device and storage medium, the method comprising: obtaining video playback features and attribute information of a video to be recommended; inputting the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score; determining a video recommendation score according to the first prediction score and the second prediction score, and recommending the video according to the video recommendation score. By adopting the above technical solution, the influence of the false association between the video duration and the model recommendation result on the video recommendation result can be eliminated.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of video data processing, and in particular to a video recommendation method, device, electronic device, and storage medium. Background Art

[0002] As an information screening tool, the recommendation system can provide information screening and push services without explicit user demand.

[0003] At present, in the video recommendation system, the video history playback data is used as the training sample of the video recommendation model, and the completion of the video playback is used as the basis for dividing the sample label. The video recommendation model trained in this way will recommend videos based on the video completion rate, resulting in deviations in the model prediction results and affecting the video recommendation effect. Summary of the invention

[0004] The embodiments of the present disclosure provide a video recommendation method, device, electronic device and storage medium, which can improve the problem of model prediction bias.

[0005] In a first aspect, an embodiment of the present disclosure provides a video recommendation method, including:

[0006] Obtaining video playback features and attribute information of the video to be recommended;

[0007] Input the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score, wherein the first prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct effect of the video duration on the model prediction results;

[0008] A video recommendation score is determined according to the first prediction score and the second prediction score, and the video recommendation is performed according to the video recommendation score.

[0009] In a second aspect, an embodiment of the present disclosure further provides a video recommendation device, the device comprising:

[0010] An information acquisition module is used to acquire video playback features and attribute information of the video to be recommended;

[0011] A prediction score determination module, used for inputting the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score, wherein the first prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct effect of the video duration on the model prediction results;

[0012] The video recommendation module is used to determine a video recommendation score according to the first prediction score and the second prediction score, and make a video recommendation according to the video recommendation score.

[0013] In a third aspect, an embodiment of the present disclosure further provides an electronic device, including:

[0014] one or more processors;

[0015] a memory for storing one or more programs,

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the video recommendation method as described in the embodiment of the present disclosure.

[0017] In a fourth aspect, the embodiments of the present disclosure further provide a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the video recommendation method as described in the embodiments of the present disclosure is implemented.

[0018] The disclosed embodiment inputs the video playback features and the attribute information of the video to be recommended into the video recommendation model to obtain a first prediction score and a second prediction score, determines the video recommendation score after eliminating the duration deviation according to the first prediction score and the second prediction score, and then recommends the video according to the video recommendation score. By adopting the above technical solution, the disclosed embodiment can eliminate the influence of the false association between the video duration and the model recommendation result on the video recommendation result, solve the problem of model prediction deviation caused by false association, enable the video recommendation model to accurately learn user preferences from the video historical playback data, and improve the video recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0020] Figure 1 A flowchart of a video recommendation method provided by an embodiment of the present disclosure;

[0021] Figure 2 A schematic diagram of the structure of a video recommendation model provided by an embodiment of the present disclosure;

[0022] Figure 3 A causal graph of a video recommendation task provided by an embodiment of the present disclosure;

[0023] Figure 4 A flowchart of another video recommendation method provided by an embodiment of the present disclosure;

[0024] Figure 5 A structural block diagram of a video recommendation device provided in an embodiment of the present disclosure;

[0025] Figure 6 A structural block diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0027] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0028] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0029] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0030] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0031] Figure 1 The flowchart of a video recommendation method provided by an embodiment of the present disclosure is shown in FIG. The method can be executed by a video recommendation device, wherein the device can be implemented by software and / or hardware and can be configured in an electronic device, typically, in a server or server cluster. The video recommendation method provided by an embodiment of the present disclosure is used in a scenario where a video is recommended. Figure 1 As shown, the video recommendation method provided in this embodiment may include:

[0032] S110: Obtain video playback features and attribute information of the video to be recommended.

[0033] In the disclosed embodiment, the video playback feature is feature information related to video playback, and the video recommendation model can determine the matching score of the video based on the video playback feature.

[0034] The attribute information is used to indicate the attributes of the video to be recommended. For example, the attribute information may include video duration and video identification features. Video duration is used to indicate the duration of the video to be recommended. Video identification features, the video identification features may be the video presentation form of the video to be recommended. For example, the video identification features include video identification (video ID), video category, and video author.

[0035] Exemplarily, video playback features, the video ID of the video to be recommended, the video category of the video to be recommended, the video author of the video to be recommended, and the video length of the video to be recommended are obtained.

[0036] S120: Input the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score.

[0037] In the disclosed embodiment, the video recommendation model may include a first video recommendation model and a second video recommendation model. The first video recommendation model is an arbitrary video recommendation model trained with video playback features, video identification features and video duration as training samples. The second video recommendation model is a neural network model trained with video duration as training samples. The first prediction score is the recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is the recommendation score of the video to be recommended output by the set video recommendation model based on the direct effects of the video duration on the model prediction results. The recommendation score is used to indicate the possibility of a user playing or liking the video to be recommended.

[0038] In some embodiments, video playback features of each video sample are obtained based on the video history playback data, and video identification features and video duration of each video sample are obtained, wherein the label of each video sample is determined based on the play completion rate (PCR) of the video sample; a first video recommendation model is trained based on the video playback features, video identification features and video duration; and a second video recommendation model is trained based on the video duration.

[0039] The video history play data may be a video play record within a set time period. For example, the video history play data includes video play information and video identification features of the played video, video play time, and video duration.

[0040] A data set of video samples required for model training is generated based on the historical video playback data. The video samples include positive samples and negative samples, and the positive samples and negative samples in the video samples are marked by labels. The labels of each video sample are determined based on the video playback completion rate. Among them, the video playback completion rate can be the smaller value of the ratio of the video playback time to the video time and 1. For example, the playback completion rate PCR can be calculated using the following formula:

[0041]

[0042] In other embodiments, the video playback features and attribute information are input into a set video recommendation model to obtain a first prediction score and a second prediction score, including: inputting the video playback features, video identification features of the video to be recommended, and the video length of the video to be recommended into the first video recommendation model to obtain a first matching score; inputting the video length of the video to be recommended into the second video recommendation model to obtain a second matching score; and determining the first prediction score and the second prediction score based on the first matching score and the second matching score.

[0043] The first matching score indicates the matching degree between the current user and the video to be recommended determined by the first video recommendation model based on the video playback characteristics of the current user combined with the video identification characteristics and video duration of the video to be recommended. The second matching score indicates the matching degree between the current user and the video to be recommended determined by the second video recommendation model based on the video duration of the video to be recommended.

[0044] Figure 2 FIG. 1 is a schematic diagram of a video recommendation model provided by an embodiment of the present disclosure. Figure 2 As shown, the video playback feature u, the video identification feature v and the video duration d are used as input parameters of the first video recommendation model, and the first video recommendation model calculates the first matching score Y based on the video playback feature u, the video identification feature v and the video duration d. mFor example, the first matching score Y can be calculated using the following formula: m :

[0045] Y m =f M (U=u,V=v,D=d)

[0046] Among them, f M It can be an arbitrary video recommendation model, U represents the video playback feature, V represents the video identification feature, D represents the video duration, u represents the video playback feature value, v represents the video identification feature value, and d represents the value of the video duration.

[0047] The video duration d is used as the input parameter of the second video recommendation model, and the second matching score Y is calculated based on the video duration d by the second video recommendation model. d .

[0048] It should be noted that the embodiments of the present disclosure do not limit the model algorithms of the first video recommendation model or the second video recommendation model. For example, the first video recommendation model may be a DeepFM model, etc. The second video recommendation model may be a residual network Res-MLP, etc.

[0049] In some embodiments, the loss function

[0050] L P =BCE(Y m,d ,y)+βBCE(Y m*d ,y)

[0051] L kl =-p(y|m,d)logp(y|m * ,d)

[0052] Among them, BCE represents the binary cross entropy loss function, β is the weight for balancing the two tasks, y is the true value, and y∈{0,1} indicates whether the video is played or liked by the user, γ is the hyperparameter for balancing the KL loss, and D represents the dataset of video samples.

[0053] Figure 3 A causal graph of a video recommendation task provided by an embodiment of the present disclosure. Figure 3As shown in the figure, a causal graph is used to describe the causal relationship in the video recommendation task, that is, the relationship between the video playback feature U, the video identification feature V, the matching score M, the video duration D and the model prediction result Y. From the causal relationship graph, it can be seen that the video duration D can affect the model prediction result Y in two ways. Specifically, D indirectly affects Y through the intermediate variable M, corresponding to the path {U, V, D}→M→Y in the causal graph. D can also have a direct impact on Y, corresponding to the path D→Y in the causal graph. This is the direct impact of the video duration on the model prediction result, which will cause the model recommendation result to deviate from the user's preference.

[0054] In some embodiments, determining a first prediction score and a second prediction score based on a first match score and a second match score includes: fusing the first match score and the second match score using a set fusion function to obtain a first prediction score; fusing the target parameter and the second match score using a set fusion function to obtain a second prediction score, wherein the target parameter is determined by inputting the video playback feature and the video identification feature as a zero vector into the first video recommendation model.

[0055] The fusion function is used to fuse the output data of the video prediction model, and the embodiment of the present disclosure does not specifically limit the specific form of the fusion function.

[0056] Optionally, after the first video recommendation model is trained according to the video playback features, video identification features and video duration, the video playback features and video identification features are input as zero vectors into the first video recommendation model to obtain a target parameter. The target parameter may be a learnable parameter. Assume that represents the target parameter, then in, represents the first video recommendation model f in the counterfactual world without (u,v) as input M The output result is:

[0057] Reference Figure 2 The structure of the video recommendation model shown in the figure can be fused into Y m and Y d Fusion is performed to obtain the first prediction score Y m,d . You can also use the fusion function to adjust the target parameters and Y d Fusion is performed to obtain the second prediction score

[0058] S130: Determine a video recommendation score according to the first prediction score and the second prediction score, and make a video recommendation according to the video recommendation score.

[0059] In the disclosed embodiment, the video recommendation score is the recommendation score of the video to be recommended after eliminating the direct influence of the video length on the video recommendation model.

[0060] In some embodiments, determining a video recommendation score based on a first prediction score and a second prediction score includes obtaining an adjustment coefficient, calculating a product of the adjustment coefficient and the second prediction score; and determining the video recommendation score based on a difference between the first prediction score and the product.

[0061] The adjustment coefficient is a hyperparameter shared across users and videos. By adjusting this coefficient, a balance between clicks and viewing time is achieved, thereby controlling the degree of eliminating duration deviation. By adjusting the adjustment coefficient in the testing phase to determine the adjustment coefficient that meets the application scenario, the adverse effects of duration deviation on model prediction results can be eliminated in actual inference scenarios.

[0062] The disclosed embodiment provides a video recommendation method, which obtains a first prediction score and a second prediction score by inputting video playback features and attribute information of the video to be recommended into a video recommendation model, determines the video recommendation score after eliminating the duration deviation according to the first prediction score and the second prediction score, and then recommends the video according to the video recommendation score. By adopting the above technical solution, the disclosed embodiment can eliminate the influence of the false association between the video duration and the model recommendation result on the video recommendation result, solve the problem of model prediction deviation caused by false association, enable the video recommendation model to accurately learn user preferences from the video historical playback data, and improve the video recommendation effect.

[0063] Currently, a user's playback time is greater than or equal to the length of the video as a completed video. If the video is completed, the video sample is determined as a positive sample, otherwise it is determined as a negative sample. Since short videos have a shorter playback time and are more likely to be completed, the completed label has a strong preference for short videos. If the video recommendation model is trained based on the completed label, the video recommendation model will increase the preference for short videos, and thus fail to truly capture the user's preference, resulting in poor model recommendation results.

[0064] To solve the above problems, the present disclosure also provides another video recommendation method. Figure 4 A flowchart of another video recommendation method provided by an embodiment of the present disclosure. The solution in this embodiment can be combined with one or more optional solutions in the above embodiments, and this embodiment adds a label determination method for video samples on the basis of the above embodiments.

[0065] like Figure 4 As shown, the video recommendation method provided in this embodiment may include:

[0066] S401. Obtain the video playback time and video duration of each video sample according to the video history playback data.

[0067] S402: For each video sample of the set duration, determine the playback completion rate according to the video playback duration and the video duration.

[0068] In the disclosed embodiment, video samples in the video sample set are screened according to the video duration to obtain sets of video samples of different durations. For example, video samples with a video duration of 20 seconds belong to one set, video samples with a video duration of 60 seconds belong to one set, video samples with a video duration of 200 seconds belong to one set, and so on.

[0069] For a 20-second video sample, the playback completion rate is determined based on the actual playback time of each video sample and the video duration. The playback completion rate PCR of a 20-second video sample can be calculated using the following formula: 20 :

[0070]

[0071] Similarly, for 60-second or 200-second video samples, the playback completion rate is determined based on the actual playback duration of each video sample and the video duration. The specific formula is as follows:

[0072]

[0073] S403: Generate a set of playback completion rates of the set of video samples of the set duration according to the playback completion rate of each video sample.

[0074] Exemplarily, the playback completion rate of each video sample in a set of video samples of different durations is calculated respectively to obtain a plurality of sets of playback completion rates at a plurality of video durations.

[0075] Using the above example, based on the PCR of each video sample in the video sample set with a video duration of 20 seconds 20 , forming a set of playback completion rates for 20-second videos. It should be noted that the number of video samples in the 20-second video sample set is consistent with the number of playback completion rates in the playback completion rate set. Similarly, according to the PCR of each video sample in the video sample set with a video duration of 60 seconds, 60 , forming a set of completion rates of 60-second videos. Based on the PCR of each video sample in the set of video samples with a video duration of 200 seconds, 200 , constituting a set of playback completion rates of 200-second videos.

[0076] S404: Determine a sample division threshold according to the playback completion rate set, and determine a label of each video sample in the video sample set according to the sample division threshold.

[0077] The sample division threshold is a division standard for determining a video sample as a positive sample or a negative sample. A positive sample may be a video played or liked by a user, and a negative sample may be a video not played or liked by a user. 1 may be used to represent a positive sample, and 0 may be used to represent a negative sample, and 1 and 0 are labels of the video sample.

[0078] In some embodiments, a sample division threshold is determined based on the playback completion rate set, and a label of each video sample in the video sample set is determined based on the sample division threshold, including: discretizing the playback completion rate set to obtain a plurality of data blocks; for each data block, dividing the playback completion rate set into two playback completion rate subsets according to the target playback completion rate in the current data block; determining the sample division threshold that makes the data distribution of the two playback completion rate subsets meet a preset condition; and determining the label of each video sample in the video sample set of the corresponding duration based on the sample division threshold and the video duration of the video sample.

[0079] It should be noted that for videos of different lengths, the playback completion rate presents a bimodal distribution. Therefore, by determining a suitable sample division threshold, can the data difference within the playback completion rate subsets divided based on the sample division threshold be minimized, and the difference between the two playback completion rate subsets be maximized. The sample division threshold can be determined by data calculation with the goal of maximizing the data difference between subsets and minimizing the data difference within subsets. For example, the variance and average standard deviation between the playback completion rate subsets can be determined; the sample division threshold can be determined based on the variance and average standard deviation corresponding to each data block.

[0080] For example, the set of playback completion rates corresponding to video samples of the same duration is discretized into S groups, each group is a data block, and the probability of each data block is p i For each data block, the maximum playback completion rate in the current data block can be determined as the target playback completion rate t; and the playback completion rate set is divided into two playback completion rate subsets: C0 and C1 according to the target playback completion rate.

[0081] Among them, the probability of C0 is w0, and The mean value is u0 and the variance is The probability of C1 is w1, and The mean is u1 and the variance is

[0082] The following formula can be used to calculate the inter-class variance of the two playback completion rate subsets C0 and C1:

[0083]

[0084] Among them, u represents the total average value of C0 and C1, and u=w0u0+w1u1, so, It can also be expressed as:

[0085]

[0086] The average standard deviation of C0 and C1 can be calculated using the following formula:

[0087]

[0088] According to the inter-class variance corresponding to each data block and the mean standard deviation Determine the sample division threshold t from each target playback completion rate * :

[0089]

[0090] It should be noted that the sample division threshold can be determined based on achieving the minimum intra-class variance and the maximum inter-class variance. According to the analysis, the distribution of the playback completion rate presents a bimodal distribution. By minimizing the average standard deviation Can make the class more compact.

[0091] After determining the sample division thresholds for video samples of different durations, the duration threshold is obtained by multiplying the sample division threshold by the corresponding video duration. The labels of video samples whose playback time is greater than the duration threshold are determined as 1, i.e., positive samples, and the labels of video samples whose playback time is less than the duration threshold are determined as 0, i.e., negative samples.

[0092] For example, for a 20s video sample, assuming its sample segmentation threshold is 0.9, the duration threshold is 18s. Only video samples with a playback duration greater than or equal to 18s will be classified as positive samples with a label of 1, while video samples with a playback duration less than 18s will be classified as negative samples with a label of 0.

[0093] Similarly, for a 60s video sample, assuming that its sample segmentation threshold is 0.8 and the calculation duration threshold is 48s, only video samples with a playback duration of 48s or longer will be classified as positive samples with a label of 1, while video samples with a playback duration of less than 48s will be classified as negative samples with a label of 0. Video samples of other durations have similar methods for determining labels, which will not be described here.

[0094] It should be noted that calculating the playback completion rates for video samples of different video lengths respectively, and then calculating the sample division threshold based on the playback completion rates, is conducive to more fair determination of the labels of the video samples.

[0095] S405. Obtain video playback features and video identification features of each video sample according to the video history playback data, wherein the label of each video sample is determined based on the playback completion rate of the video sample.

[0096] For example, let the video playback feature dataset be {u1,u2,...,u N}, the video identification feature dataset is {v1,v2,…,v M}, the video duration dataset is {d1,d2,…,d L}, where N represents the number of users, M represents the number of videos, and L represents the duration level. Then the video sample dataset can be expressed as {(u,v,d,y)|u∈U,v∈V,d∈D}, where y is the true value, that is, the label of the video sample, and y∈{0,1}.

[0097] S406: Train the first video recommendation model according to the video playback features, video identification features, and video duration.

[0098] S407: Training the second video recommendation model according to the video duration.

[0099] S408: Obtain video playback features and attribute information of the video to be recommended, wherein the attribute information includes video duration and video identification features.

[0100] S409: Input the video playback feature, the video identification feature of the video to be recommended, and the video duration of the video to be recommended into a first video recommendation model to obtain a first matching score.

[0101] S410: Input the video duration of the video to be recommended into a second video recommendation model to obtain a second matching score.

[0102] S411. Determine a first prediction score and a second prediction score according to the first matching score and the second matching score.

[0103] S412: Determine a video recommendation score according to the first prediction score and the second prediction score, and make a video recommendation according to the video recommendation score.

[0104] In some embodiments, model testing is required after model training. Only after the model test passes can it be used to predict the possibility of interaction between the user and the video to be recommended. The embodiments of the present disclosure do not specifically limit the model testing. It should be noted that in the video recommendation model training stage, the video recommendation model has two output results Y m,d and Y m*d , in the model testing phase, through Y m,d and Y m*dThe difference between and determines the video recommendation score.

[0105] This embodiment provides a video recommendation method. Before training a video recommendation model, the label of a video sample is determined by the playback completion rate, which reduces the preference of the completion label for short videos at the data level, and uses the label of the video sample determined based on the playback completion rate as the basic label for training the video recommendation model, thereby eliminating the influence of the duration deviation on the model prediction result, accurately capturing the user's preference, and avoiding the problem that the video recommendation model trained with the completion as the basic label will aggravate the preference for short videos, thereby further improving the accuracy of video recommendation.

[0106] Figure 5 This is a structural block diagram of a video recommendation device provided by an embodiment of the present disclosure. The device can be implemented by software and / or hardware and can be configured in an electronic device, typically, in a server or server cluster. Figure 5 As shown, the video recommendation device provided in this embodiment may include: an information acquisition module 510, a prediction score determination module 520, and a video recommendation module 530, wherein:

[0107] The information acquisition module 510 is used to acquire video playback features and attribute information of the video to be recommended;

[0108] A prediction score determination module 520 is used to input the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score, wherein the first prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct effect of the video duration on the model prediction results;

[0109] The video recommendation module 530 is configured to determine a video recommendation score according to the first prediction score and the second prediction score, and perform video recommendation according to the video recommendation score.

[0110] The disclosed embodiment provides a video recommendation device, which obtains a first prediction score and a second prediction score by inputting video playback features and attribute information of the video to be recommended into a video recommendation model, determines the video recommendation score after eliminating the duration deviation according to the first prediction score and the second prediction score, and then recommends the video according to the video recommendation score. By adopting the above technical solution, the disclosed embodiment can eliminate the influence of the false association between the video duration and the model recommendation result on the video recommendation result, solve the problem of model prediction deviation caused by false association, enable the video recommendation model to accurately learn user preferences from the video historical playback data, and improve the video recommendation effect.

[0111] In the above solution, the set video recommendation model may include a first video recommendation model and a second video recommendation model; the device further includes:

[0112] A historical data acquisition module, used to acquire video playback features of each video sample according to the video historical playback data, and to acquire video identification features and video duration of each video sample, wherein the label of each video sample is determined based on the playback completion rate of the video sample;

[0113] A first model training module, used for training the first video recommendation model according to the video playback features, video identification features and video duration;

[0114] The second model training module is used to train the second video recommendation model according to the video length.

[0115] Furthermore, the prediction score determination module 520 is specifically used to:

[0116] Inputting the video playback feature, the video identification feature of the video to be recommended, and the video duration of the video to be recommended into a first video recommendation model to obtain a first matching score;

[0117] Inputting the video duration of the video to be recommended into a second video recommendation model to obtain a second matching score;

[0118] The first prediction score and the second prediction score are determined based on the first matching score and the second matching score.

[0119] In the above scheme, the device also includes:

[0120] A playback duration determination module is used to obtain the video playback duration of each video sample based on the video history playback data;

[0121] A playback completion rate determination module, for determining the playback completion rate of each video sample of a set duration according to the video playback duration and the video duration;

[0122] A set generation module, used to generate a set of playback completion rates of the video sample set of the set duration according to the playback completion rate of each video sample;

[0123] The label determination module is used to determine a sample division threshold according to the playback completion rate set, and determine a label of each video sample in the video sample set according to the sample division threshold.

[0124] Furthermore, the tag determination module is specifically used for:

[0125] Discretizing the playback completion rate set to obtain a plurality of data blocks;

[0126] For each data block, dividing the play completion rate set into two play completion rate subsets according to the target play completion rate in the current data block;

[0127] Determine the sample division threshold value that makes the data distribution of the two playback completion rate subsets meet a preset condition;

[0128] According to the sample division threshold and the video duration of the video sample, a label of each video sample in a video sample set of corresponding duration is determined.

[0129] Furthermore, the prediction score determination module 520 is further specifically configured to:

[0130] Using a set fusion function to fuse the first matching score and the second matching score to obtain a first prediction score;

[0131] A set fusion function is used to fuse the target parameter and the second matching score to obtain a second prediction score, wherein the target parameter is determined by inputting the video playback feature and the video identification feature as a zero vector into the first video recommendation model.

[0132] In the above solution, the video recommendation module 530 is specifically used for:

[0133] Obtaining an adjustment coefficient, and calculating a product of the adjustment coefficient and a second prediction score;

[0134] The video recommendation score is determined according to a difference between the first prediction score and the product.

[0135] The video recommendation device provided in the embodiment of the present disclosure can execute the video recommendation method provided in any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of executing the video recommendation method. For technical details not described in detail in this embodiment, please refer to the video recommendation method provided in any embodiment of the present disclosure.

[0136] Figure 6 This is a structural block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 6 , which shows the structure of an electronic device 600 suitable for implementing the embodiment of the present disclosure. The terminal device in the embodiment of the present disclosure may include but is not limited to laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0137] like Figure 6As shown, the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0138] Typically, the following devices may be connected to the I / O interface 605: an input device 606 and an output device 607; a storage device 608 such as a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Figure 6 The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0139] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0140] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0141] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0142] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0143] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains video playback features and attribute information of the video to be recommended;

[0144] Input the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score, wherein the first prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct effect of the video duration on the model prediction results;

[0145] A video recommendation score is determined according to the first prediction score and the second prediction score, and the video recommendation is performed according to the video recommendation score.

[0146] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0147] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0148] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a module does not, in some cases, constitute a limitation on the unit itself.

[0149] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0150] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0151] According to one or more embodiments of the present disclosure, Example 1 provides a video recommendation method, including:

[0152] Obtaining video playback features and attribute information of the video to be recommended;

[0153] Input the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score, wherein the first prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct effect of the video duration on the model prediction results;

[0154] A video recommendation score is determined according to the first prediction score and the second prediction score, and the video recommendation is performed according to the video recommendation score.

[0155] According to one or more embodiments of the present disclosure, Example 2 According to the method described in Example 1, the set video recommendation model includes a first video recommendation model and a second video recommendation model;

[0156] Obtaining video playback features of each video sample according to the video history playback data, and obtaining video identification features and video duration of each video sample, wherein the label of each video sample is determined based on the playback completion rate of the video sample;

[0157] Training the first video recommendation model according to the video playback features, video identification features, and video duration;

[0158] The second video recommendation model is trained according to the video duration.

[0159] According to one or more embodiments of the present disclosure, Example 3 is a method according to Example 2, wherein the step of inputting the video playback feature and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score includes:

[0160] Inputting the video playback feature, the video identification feature of the video to be recommended, and the video duration of the video to be recommended into a first video recommendation model to obtain a first matching score;

[0161] Inputting the video duration of the video to be recommended into a second video recommendation model to obtain a second matching score;

[0162] The first prediction score and the second prediction score are determined based on the first matching score and the second matching score.

[0163] According to one or more embodiments of the present disclosure, Example 4 is the method according to Example 2, further comprising:

[0164] Obtain the video playback duration of each video sample according to the video historical playback data;

[0165] For each video sample of a set duration, determining a playback completion rate according to the video playback duration and the video duration;

[0166] Generating a set of playback completion rates of the set of video samples of the set duration according to the playback completion rate of each video sample;

[0167] A sample division threshold is determined according to the playback completion rate set, and a label of each video sample in the video sample set is determined according to the sample division threshold.

[0168] According to one or more embodiments of the present disclosure, Example 5 is the method according to Example 4, wherein determining a sample division threshold according to the playback completion rate set, and determining a label of each video sample in the video sample set according to the sample division threshold includes:

[0169] Discretizing the playback completion rate set to obtain a plurality of data blocks;

[0170] For each data block, dividing the play completion rate set into two play completion rate subsets according to the target play completion rate in the current data block;

[0171] Determine the sample division threshold value that makes the data distribution of the two playback completion rate subsets meet a preset condition;

[0172] According to the sample division threshold and the video duration of the video sample, a label of each video sample in a video sample set of corresponding duration is determined.

[0173] According to one or more embodiments of the present disclosure, Example 6 is the method according to Example 3, wherein determining the first prediction score and the second prediction score according to the first matching score and the second matching score includes:

[0174] Using a set fusion function to fuse the first matching score and the second matching score to obtain a first prediction score;

[0175] A set fusion function is used to fuse the target parameter and the second matching score to obtain a second prediction score, wherein the target parameter is determined by inputting the video playback feature and the video identification feature as a zero vector into the first video recommendation model.

[0176] According to one or more embodiments of the present disclosure, Example 7 is the method according to Example 1, wherein determining the video recommendation score according to the first prediction score and the second prediction score includes:

[0177] Obtaining an adjustment coefficient, and calculating a product of the adjustment coefficient and a second prediction score;

[0178] The video recommendation score is determined according to a difference between the first prediction score and the product.

[0179] According to one or more embodiments of the present disclosure, Example 8 provides a video recommendation device, including:

[0180] An information acquisition module is used to acquire video playback features and attribute information of the video to be recommended;

[0181] A prediction score determination module, used for inputting the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score, wherein the first prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct effect of the video duration on the model prediction results;

[0182] The video recommendation module is used to determine a video recommendation score according to the first prediction score and the second prediction score, and make a video recommendation according to the video recommendation score.

[0183] According to one or more embodiments of the present disclosure, Example 9 provides an electronic device, including:

[0184] one or more processors;

[0185] a memory for storing one or more programs,

[0186] When the one or more programs are executed by the one or more processors, the one or more processors implement the video recommendation method as described in any one of Examples 1-7.

[0187] According to one or more embodiments of the present disclosure, Example 10 provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the video recommendation method as described in any one of Examples 1-7.

[0188] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0189] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0190] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A video recommendation method, characterized in that: include: Acquire video playback features and attribute information of the video to be recommended, wherein the attribute information includes video identification features and video duration; Inputting the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score, wherein the first prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct effect of the video duration on the model prediction results, and the set video recommendation model includes a first video recommendation model and a second video recommendation model; Acquire an adjustment coefficient, calculate a product of the adjustment coefficient and a second prediction score, determine a video recommendation score according to a difference between the first prediction score and the product, and make a video recommendation according to the video recommendation score; The step of inputting the video playback feature and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score includes: Inputting the video playback feature, the video identification feature of the video to be recommended, and the video duration of the video to be recommended into a first video recommendation model to obtain a first matching score; Inputting the video duration of the video to be recommended into a second video recommendation model to obtain a second matching score; Using a set fusion function to fuse the first matching score and the second matching score to obtain a first prediction score; A set fusion function is used to fuse the target parameter and the second matching score to obtain a second prediction score, wherein the target parameter is determined by inputting the video playback feature and the video identification feature as a zero vector into the first video recommendation model.

2. The method according to claim 1, characterized in that The set video recommendation model includes a first video recommendation model and a second video recommendation model; Obtaining video playback features of each video sample according to the video history playback data, and obtaining video identification features and video duration of each video sample, wherein the label of each video sample is determined based on the playback completion rate of the video sample; Training the first video recommendation model according to the video playback features, video identification features, and video duration; The second video recommendation model is trained according to the video duration.

3. The method according to claim 2, characterized in that Also includes: Obtain the video playback duration of each video sample according to the video historical playback data; For each video sample of a set duration, determining a playback completion rate according to the video playback duration and the video duration; Generating a set of playback completion rates of the set of video samples of the set duration according to the playback completion rate of each video sample; A sample division threshold is determined according to the playback completion rate set, and a label of each video sample in the video sample set is determined according to the sample division threshold.

4. The method according to claim 3, characterized in that The step of determining a sample division threshold according to the playback completion rate set, and determining a label of each video sample in the video sample set according to the sample division threshold, includes: Discretizing the playback completion rate set to obtain a plurality of data blocks; For each data block, dividing the play completion rate set into two play completion rate subsets according to the target play completion rate in the current data block; Determine the sample division threshold value that makes the data distribution of the two playback completion rate subsets meet a preset condition; According to the sample division threshold and the video duration of the video sample, a label of each video sample in a video sample set of corresponding duration is determined.

5. A video recommendation device, characterized in that: include: An information acquisition module, used to acquire video playback features and attribute information of the video to be recommended, wherein the attribute information includes video identification features and video duration; A prediction score determination module, used for inputting the video playback features and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score, wherein the first prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct and indirect effects of the video duration on the model prediction results, and the second prediction score is a recommendation score of the video to be recommended output by the set video recommendation model based on the direct effect of the video duration on the model prediction results, and the set video recommendation model includes a first video recommendation model and a second video recommendation model; A video recommendation module, configured to obtain an adjustment coefficient, calculate a product of the adjustment coefficient and a second prediction score, determine a video recommendation score according to a difference between the first prediction score and the product, and make a video recommendation according to the video recommendation score; The step of inputting the video playback feature and attribute information into a set video recommendation model to obtain a first prediction score and a second prediction score includes: Inputting the video playback feature, the video identification feature of the video to be recommended, and the video duration of the video to be recommended into a first video recommendation model to obtain a first matching score; Inputting the video duration of the video to be recommended into a second video recommendation model to obtain a second matching score; Using a set fusion function to fuse the first matching score and the second matching score to obtain a first prediction score; A set fusion function is used to fuse the target parameter and the second matching score to obtain a second prediction score, wherein the target parameter is determined by inputting the video playback feature and the video identification feature as a zero vector into the first video recommendation model.

6. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the video recommendation method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the video recommendation method as described in any one of claims 1 to 4 is implemented.

Images