CN116366923A - Video recommendation method and device and electronic equipment - Google Patents
Video recommendation method and device and electronic equipment Download PDFInfo
- Publication number
- CN116366923A CN116366923A CN202111613794.5A CN202111613794A CN116366923A CN 116366923 A CN116366923 A CN 116366923A CN 202111613794 A CN202111613794 A CN 202111613794A CN 116366923 A CN116366923 A CN 116366923A
- Authority
- CN
- China
- Prior art keywords
- video
- interest
- model
- recommendation
- candidate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 239000013598 vector Substances 0.000 claims abstract description 141
- 238000009826 distribution Methods 0.000 claims abstract description 103
- 239000011159 matrix material Substances 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims description 36
- 230000006870 function Effects 0.000 claims description 19
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 abstract description 11
- 230000008569 process Effects 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 238000009331 sowing Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
- H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
- H04N21/4668—Learning process for intelligent management, e.g. learning user preferences for recommending movies for recommending content, e.g. movies
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/73—Querying
- G06F16/735—Filtering based on additional data, e.g. user or group profiles
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
- H04N21/258—Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data
- H04N21/25866—Management of end-user data
- H04N21/25891—Management of end-user data being end-user preferences
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Theoretical Computer Science (AREA)
- Computer Graphics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
Abstract
The application provides a video recommendation method, a video recommendation device and electronic equipment, wherein the method comprises the following steps: acquiring a video sequence, wherein the video sequence comprises a plurality of videos and playing time of each video; inputting target time and the video sequence into a first model to predict an interest distribution vector so as to obtain the interest distribution vector corresponding to the target time, wherein the interest distribution vector is used for representing weight distribution of multidimensional interests; inputting a history video set, a plurality of candidate videos and a multidimensional interest vector matrix which are acquired in advance into a second model to predict index parameters so as to acquire the predicted index parameters of each candidate video; acquiring recommendation scores of each candidate video based on the interest distribution vector and the predictor parameters; and recommending the plurality of candidate videos according to the recommendation score of each candidate video. The video recommendation method and device can improve the video recommendation effect.
Description
Technical Field
The present disclosure relates to the field of network transmission technologies, and in particular, to a video recommendation method and apparatus, and an electronic device.
Background
The basis of the existing target recommendation technical scheme is recommendation aiming at a single user, but in an internet home television scene, the user composition is complex, the user is often a family, a plurality of users in the family have the conditions of a plurality of age groups and a large interest point span, the recommendation is performed by taking a terminal as a unit in the home television scene, the currently watched user cannot be known, and if video recommendation is performed only based on a certain recommendation target, the recommendation effect is poor.
Disclosure of Invention
The application provides a video recommending method, a video recommending device and electronic equipment, and aims to solve the problem of poor video recommending effect.
In a first aspect, an embodiment of the present application provides a video recommendation method, including:
acquiring a video sequence, wherein the video sequence comprises a plurality of videos and playing time of each video;
inputting target time and the video sequence into a first model to predict an interest distribution vector so as to obtain the interest distribution vector corresponding to the target time, wherein the interest distribution vector is used for representing weight distribution of multidimensional interests;
inputting a history video set, a plurality of candidate videos and a multidimensional interest vector matrix which are acquired in advance into a second model to predict index parameters so as to acquire the predicted index parameters of each candidate video;
acquiring recommendation scores of each candidate video based on the interest distribution vector and the predictor parameters;
and recommending the plurality of candidate videos according to the recommendation score of each candidate video.
In a second aspect, an embodiment of the present application further provides a video recommendation apparatus, including:
the first acquisition module is used for acquiring a video sequence, wherein the video sequence comprises a plurality of videos and the playing time of each video;
the first prediction module is used for inputting the target time and the video sequence into a first model to predict an interest distribution vector so as to obtain the interest distribution vector corresponding to the target time, wherein the interest distribution vector is used for representing weight distribution of multidimensional interests;
the second prediction module is used for inputting a history video set, a plurality of candidate videos and a multidimensional interest vector matrix which are acquired in advance into a second model to predict index parameters so as to acquire the prediction index parameters of each candidate video;
the second acquisition module is used for acquiring the recommendation score of each candidate video based on the interest distribution vector and the prediction index parameter;
and the recommending module is used for recommending the plurality of candidate videos according to the recommending score of each candidate video.
In a third aspect, embodiments of the present application further provide an electronic device, including: a transceiver, a memory, a processor, and a program stored on the memory and executable on the processor; the processor is configured to read a program in the memory to implement the steps in the method according to the first aspect of the embodiment of the present application.
In a fourth aspect, embodiments of the present application further provide a readable storage medium having a program stored thereon, which when executed by a processor, implements the steps of the method according to the first aspect of embodiments of the present application.
In the embodiment of the application, the target time and the video sequence are input into a first model to predict an interest distribution vector so as to obtain the interest distribution vector corresponding to the target time, wherein the interest distribution vector is used for representing weight distribution of multidimensional interests; inputting a history video set, a plurality of candidate videos and a multidimensional interest vector matrix which are acquired in advance into a second model to predict index parameters so as to acquire the predicted index parameters of each candidate video; acquiring recommendation scores of each candidate video based on the interest distribution vector and the predictor parameters; the recommendation score of each candidate video can be determined based on the predicted weight distribution of the multi-dimensional interests and the predicted index parameters, so that video recommendation is performed by combining the multi-interests and the recommendation targets, and the effect of video recommendation is improved.
Drawings
In order to more clearly illustrate the technical solutions of the present application, the following description will briefly introduce the drawings that are needed in the embodiments or the description of the prior art, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort to a person skilled in the art.
Fig. 1 is a schematic flow chart of a video recommendation method according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of a video recommendation device according to an embodiment of the present application;
fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
The terms "first," "second," and the like in embodiments of the present application are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus. Furthermore, the use of "and/or" in this application means at least one of the connected objects, such as a and/or B and/or C, is meant to encompass the 7 cases of a alone, B alone, C alone, and both a and B, both B and C, both a and C, and both A, B and C.
Referring to fig. 1, fig. 1 is a flowchart of a video recommendation method provided in an embodiment of the present application, as shown in fig. 1, including the following steps:
It will be appreciated that in the video sequence, the identification numbers (Identity Document, ID) of the videos may be used to characterize different videos, and that the video sequence may be a video viewing record acquired correspondingly during a period of time preceding the current time, the video viewed at each time being represented by the form of the video sequence. The plurality of videos of the video sequence may include the same video, for example: under the condition that videos corresponding to a plurality of moments are acquired at preset time intervals, if the duration of the watched videos is longer, the videos corresponding to adjacent moments may be consistent; or it is possible to watch the same video again at a different time, such as: different users of the home television watch the same video at different times, the same user watches the same video repeatedly, etc.
In addition, the playing time of each video in the video sequence may be represented by the starting time of the video viewing or the median of the viewing time period, or the playing time of each video recorded in the video sequence may be equally spaced, and the video sequence may be composed by acquiring the video corresponding to each time interval.
Wherein the multi-dimensional interest may be predetermined, for example: multiple interest categories may be extracted by capturing a set of historically viewed videos, each dimension of interest may correspond to one interest category, for example: interests may be categorized in terms of the presentation of the video as: movie interests, cartoon interests, variety interests, etc.; alternatively, interests may be categorized by the content of the video as: food interest, fun interest, popular science interest, game interest, and the like.
The interest distribution vector is used for representing weight distribution of multidimensional interests, namely, the preference of the user for interests of each dimension under the target time can be predicted through the first model.
Optionally, before the target time and the video sequence are input into the first model to predict the interest distribution vector in step 102, the method may further include the following steps:
acquiring the historical video set;
and extracting features of the historical video set by using a third model to obtain the multidimensional interest vector matrix, wherein the multidimensional interest vector matrix comprises a plurality of interest vectors, and each interest vector is used for representing one-dimensional interest.
Specifically, the third model may be constructed based on a multi-interest serialization recommendation (Controllable Multi-Interest Framework for Recommendation, comiRec) algorithm, and the multi-dimensional interest vector matrix is formed by capturing a historical video set, capturing multi-dimensional interests, and characterizing each of the multi-dimensional interests by using an interest vector.
In some embodiments, for example: in a home television scene, historical viewing data of each home user in a period of time can be obtained according to a preset time interval to extract a historical video set, so that the third model is used for extracting features of the historical video set to obtain a multidimensional interest vector matrix corresponding to the period of time.
In this embodiment, feature extraction is performed on the historical video set by using a third model to obtain the multidimensional interest vector matrix, that is, decoupling of multidimensional interests is implemented through the historical video set, that is, the existing multidimensional interests are determined based on the historical video set.
It may be appreciated that the target time may be a time for performing video recommendation, and the target time and the video sequence are input into the first model to predict a weight distribution of the multi-dimensional interests under the target time, that is, a sum of weights of the multi-dimensional interests is 1, that is, a modular length of the interest distribution vector is 1.
And 103, inputting the pre-acquired historical video set, the plurality of candidate videos and the multi-dimensional interest vector matrix into a second model to predict index parameters so as to acquire the predicted index parameters of each candidate video.
The candidate videos may be all videos in a resource library, or may be videos screened in advance based on the historical video set. The prediction index parameter of each candidate video may include a prediction index parameter corresponding to each dimension of interest, that is, an index parameter of each category of interest for the candidate video.
The index parameters may include measurement parameters of various targets, for example: click rate, play rate, viewing duration, increment income, etc., can be used to measure the corresponding target realization effect, respectively.
And 104, acquiring a recommendation score of each candidate video based on the interest distribution vector and the prediction index parameter.
It is understood that the recommendation score is obtained based on the interest distribution vector and the prediction index parameter, i.e. the recommendation score may comprehensively consider the multidimensional interest and index parameter, and not just make video recommendation for improving a certain index.
Optionally, in step 104, the obtaining a recommendation score of each candidate video based on the interest distribution vector and the prediction parameter information may specifically include:
acquiring recommendation scores of each candidate video by using a recommendation function based on the interest distribution vector and the predictor parameters;
the recommendation function comprises a first equalization parameter, wherein the first equalization parameter is used for equalizing the multidimensional interest and the prediction index parameter.
The interest distribution vector may represent a category distribution probability of the recommended video at the target time, for example: the video category corresponding to the interest vector with the greatest weight in the interest distribution vector can represent the video category most likely to be of interest to the user in the target time. The corresponding index realization effect can be represented by the prediction index parameter, and the improvement of the index realization effect can be realized under the condition of ensuring user experience by the first equalization parameter, so that the multidimensional interest and the prediction index parameter are equalized.
In this embodiment, a recommendation function is used to obtain a recommendation score of each candidate video based on the interest distribution vector and the predictor parameter, where the recommendation function includes a first equalization parameter, where the first equalization parameter is used to equalize the multidimensional interest and the predictor parameter, that is, the multidimensional interest and the predictor parameter can be equalized by a first mean parameter in the recommendation function, and relative equalization of the multidimensional interest and the predictor parameter can be achieved by adjusting the first mean parameter.
The execution main body of the embodiment of the application can be a server or video display terminal equipment (such as a television) or other terminal equipment, and when the execution main body is the video display terminal equipment, the video can be directly displayed according to the recommendation; under the condition that the execution subject is a server or other terminal equipment, the video recommendation result can be sent to the video display terminal equipment, and then the corresponding video can be displayed through the video display terminal equipment.
In the embodiment of the application, the target time and the video sequence are input into a first model to predict an interest distribution vector so as to obtain the interest distribution vector corresponding to the target time, wherein the interest distribution vector is used for representing weight distribution of multidimensional interests; inputting a history video set, a plurality of candidate videos and a multidimensional interest vector matrix which are acquired in advance into a second model to predict index parameters so as to acquire the predicted index parameters of each candidate video; acquiring recommendation scores of each candidate video based on the interest distribution vector and the predictor parameters; the recommendation score of each candidate video can be determined based on the predicted weight distribution of the multi-dimensional interests and the predicted index parameters, so that video recommendation is performed by combining the multi-interests and the recommendation targets, and the effect of video recommendation is improved.
Optionally, the predictor parameters include click rate information and play completion rate information corresponding to each dimension of interest;
the recommendation function further comprises a second equalization parameter, wherein the second equalization parameter is used for adjusting the relative weight of the click rate information and the play completion rate information.
In some embodiments, the predictor parameter of each candidate video may be obtained by weighted summation of the click rate information and the finish rate information, and the second equalization parameter may be understood as a weight of any one of the click rate and the finish rate, and the corresponding weight of the other one is 1-second equalization parameter.
In this embodiment, the recommendation function further includes a second equalization parameter, where the second equalization parameter is used to adjust a relative weight of the click rate information and the play completion rate information, that is, equalization of the relative weight of the click rate information and the play completion rate information may be achieved through the second equalization parameter in the recommendation function, and adjustment of the relative weight of the click rate information and the play completion rate information may be achieved through adjustment of the second equalization parameter.
Optionally, the second model is trained by:
inputting a first training sample to the second model to obtain output of the second model, wherein the first training sample comprises a historical video set sample, the multidimensional interest vector matrix and click rate labels and play completion rate labels of each video in the historical video set sample;
based on a Bayesian personalized ordering algorithm, acquiring a first loss value of the output of the second model;
and updating model parameters of the second model based on the first loss value.
The first training samples may include negative samples, for example: multiple videos may be randomly selected as negative examples of click rate labels.
In this embodiment, the accuracy of the prediction of the second model may be improved by updating the model parameters of the second model through iterative training of the second model.
Optionally, the first model is trained by:
inputting a second training sample to the first model to obtain an output of the first model, the second training sample comprising a first time, a video sequence sample, and an interest distribution vector sample;
determining a second loss value of the output of the first model based on the interest distribution vector samples;
and updating model parameters of the first model based on the second loss value.
In this embodiment, the model parameters of the first model are updated by iterative training of the first model, so as to improve the prediction accuracy of the first model.
Optionally, the video training sample includes a target video corresponding to the first time;
the interest distribution vector samples are determined based on similarity of the target video to the multi-dimensional interest vector matrix.
The similarity between the target video and the multidimensional interest vector matrix may be determined using cosine similarity, for example, the target video may be represented as a video vector according to a plurality of videos in the video sequence, and the weight of each dimension of interest vector may be determined as the interest distribution vector sample by calculating cosine similarity between the video vector and the multidimensional interest vector matrix.
In this embodiment, by training the first model using the interest distribution vector samples determined based on the similarity between the target video and the multidimensional interest vector matrix, the interest distribution vector predicted by the first model and the interest distribution vector samples determined based on the target video tend to be consistent, and the prediction accuracy of the first model can be improved.
The various optional implementations described in the embodiments of the present application may be implemented in combination with each other without collision with each other, or may be implemented separately, which is not limited to the embodiments of the present application.
For ease of understanding, specific examples are as follows:
the embodiment of the application also provides a multi-target recommendation method which can be used for a home user scene facing the Internet television, and specifically comprises the following steps:
step 1, constructing a user interest decoupling model, and obtaining a multidimensional interest matrix V of a household user according to user watching records u Representation vector I of each article j ；
The step 1 may specifically include the following steps:
collecting a behavior history S of a user u And related characteristic data relating to the item, wherein S u ＝[I 1 ，I 2 ……，I n ]；
Using a user interest decoupling model, a home user is modeled as K interest vectors, each representing a dimension of interest, such as movie interest, cartoon interest, television show interest, and the like. The application takes a ComiRec algorithm as an example, and uses a self-attention mechanism to realize multi-interest decoupling of a user.
The multidimensional interest vector matrix V of the user u can be obtained by using the ComiRec algorithm u Representation vector I of each article j Wherein V is u For a matrix of K vectors, i.e
Wherein (1)>
The prediction process of the multi-interest decoupling model is as follows: inputting a history S of a user u I.e. the historical viewing sequence of user u, the model output is the multidimensional interest vector matrix V of the user u Wherein V is u A matrix of K vectors.
Step 2: constructing a user interest intention prediction model to realize real-time sequence H of the user according to recommended time t u The user interest distribution at time t is predicted.
The step 2 may specifically include the following steps:
building a user intention prediction model: on the basis of the step 1, a user intention prediction model is constructed, and the input of the model is the recommended time t and the history sequence H of the user u The output of the model is the K-dimensional vector d target And (2) corresponding to the weights of the K vectors after decoupling in the step (1), namely predicting the distribution of the K interests of the user at the time t.
Wherein H is u Video watched by user u at each time instant for user sequence containing historical time
H u ＝[(i 1 ，t 1 )，(i 2 ，t 2 )，……(i n ，t n )]；
Training samples of the user intent prediction model were constructed as follows: on the basis of the step 1, a K-dimensional interest matrix V of the family user is obtained u Representation vector I of each article j For user u, the historical sequence of training samples is H u Training object article is I target Training target interaction time is t target Thus, the user target item I can be calculated target K-dimensional interest with a user
Similarity between users t target Interest distribution value d at time target . A typical similarity calculation is cosine similarity, and after vector normalization (vector modulo length is 1), the calculation is as follows:
d target ＝I target *V u ；
i.e. d target ∈R K Also need to be done for d after calculation target Normalization is performed to make the sum of elements be 1, and the normalization mode is provided as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
representing the normalized interest distribution vector of the target item, < >>
Interest distribution vector representing target item, +.>
An interest distribution vector representing a target item of target j;
obtaining normalized d target D for subsequent use target The normalized result is obtained;
the input of the user intention prediction model is the history of the user H u The recommended time t is output as d predict The weight distribution of the K interests of the user predicted by the model is obtained. During training, d generated by the steps is used target As a training label, the training target is to make the interest distribution vector predicted by the model coincide with the interest distribution vector of the target object calculated by the following way, and a typical training loss is KL divergence (Kullback-Leibler divergence, KL diversity):
L KL ＝∑d target *log(d predict )；
wherein L is KL Represents the loss value, d target Representing a calculated interest distribution vector, d, of the target object predict Representing a model predicted interest distribution vector;
in the on-line inference prediction, the history H of the user is input u Outputting the predicted user interest distribution d at the current moment at the current recommended time t predict A weight distribution representing K interest distributions for the user;
step 3: constructing a multi-target recommendation model, and predicting indexes such as click rate, conversion rate, play duration and the like of candidate articles according to interest vectors of users;
step 3 may specifically include the following steps:
the input of the multi-objective recommendation model is a userHistory S of (2) u User interest vector V u Related features of the item, candidate item C i Outputting scores for each dimension interest of the user for each target of the candidate item
Where j represents a target such as click rate (click), complete rate (complete). One construction mode is to share parameters of a bottom layer, model upper layer targets respectively, and model output scoring values of the targets respectively. For convenience of explanation, the present solution uses two targets, i.e. click rate and play rate, as cases, and practical application is not limited to these two targets.
Taking click rate and play rate as examples, a training sample of the multi-target model is constructed. User history sequence S u User interest vector V u The item clicked after the sequence is I click And if the user clicks the object, the playing rate label is 1, otherwise, the playing rate label is 0. Within the range of all items, N items are randomly selected as negative examples of click tags.
Training and predicting process of multi-target model: in the training process, the output of the model is the click rate p of interest vector of the user on candidate articles click Rate p of complete sowing complete . The training target of the model is that the click rate and the completion rate of output are consistent with the actual label, and the loss function used in the application is BPR loss (Bayesian Personalized Ranking loss, bayesian personalized ordering algorithm loss);
step 4: and combining the user interest distribution at the moment t with the multi-objective prediction indexes of each interest, comprehensively considering a plurality of objectives of the interests, and sequencing candidate articles according to the comprehensive indexes to obtain a final recommendation result.
Step 4 may specifically include the following steps:
based on the above steps, for a given user u, its history sequence S is known u ，H u Candidate item C at current time t i The K-dimensional interest V of the user u can be obtained through the user interest decoupling model u By means ofThe user intention prediction model obtains the intention d of the user at the moment t, and the object C of each dimension interest of the user u is obtained through the multi-target model i Predicted click rate value of (2)
And the prediction value of the complete broadcast rate->
In the application, multiple interests and multiple targets can be balanced according to actual demands, so that the balance of multiple targets is realized on the premise of not damaging user experience, and the realization of the multiple targets is considered and ensured. The application proposes a balancing mode which is to balance interest expression and multi-objective realization effect by setting parameters m and q for scoring. The specific calculation mode is as follows:
wherein Score represents a recommendation Score, K represents a dimension of multidimensional interest, d k Representing an interest distribution vector corresponding to the interest in the k dimension, d j An interest distribution vector corresponding to the j-dimensional interest is represented, m and q represent setting parameters,
representing k-dimensional interest for item C i Click rate prediction value of->
Representing k-dimensional interest for item C i Is a predicted value of the completion rate.
The balance of multiple interests and multiple targets can be realized by controlling the parameter m, and the larger the parameter m is, the smaller the influence of the multiple interests on the multiple targets is, namely the perception capability of the multiple target recommendation on the interests is weakened. The parameter q can realize the balance among a plurality of targets, when q takes 0.5, the click rate is the same as the weight of the index of the complete sowing rate, and when q takes 1, only the index of the click rate is considered as the basis of sequencing;
and sorting the candidate articles according to the recommended scores obtained in the steps, wherein the articles with higher scores are ranked at the forefront, so that a final sorting result is obtained.
According to the method and the device for achieving the multi-target and multi-interest combination, the multi-interest targets can be comprehensively considered according to the interest preference of the current user under the recommendation scene of the home user, and recommendation experience of the home user can be improved on the basis of meeting the plurality of recommendation targets. In addition, in the calculation formula of the recommendation score, the controllable and configurable interest expression and multi-target weight can be realized by setting the parameter m and the parameter q, and the recommendation score is more flexible in the recommendation process so as to meet various recommendation requirements.
Referring to fig. 2, fig. 2 is a schematic structural diagram of a video recommendation device according to an embodiment of the present application. As shown in fig. 2, the video recommendation apparatus 200 includes:
a first obtaining module 201, configured to obtain a video sequence, where the video sequence includes a plurality of videos and a playing time of each video;
a first prediction module 202, configured to input a target time and the video sequence into a first model to predict an interest distribution vector, so as to obtain an interest distribution vector corresponding to the target time, where the interest distribution vector is used to represent weight distribution of a multidimensional interest;
the second prediction module 203 is configured to input a previously acquired historical video set, a plurality of candidate videos, and a multi-dimensional interest vector matrix into a second model to predict an index parameter, so as to acquire a prediction index parameter of each candidate video;
a second obtaining module 204, configured to obtain a recommendation score of each candidate video based on the interest distribution vector and the predictor parameters;
a recommending module 205, configured to recommend the plurality of candidate videos according to a recommendation score of each candidate video.
Optionally, the video recommendation device 200 may further include:
the third acquisition module is used for acquiring the historical video set;
and the extraction module is used for extracting features of the historical video set by using a third model to obtain the multidimensional interest vector matrix, wherein the multidimensional interest vector matrix comprises a plurality of interest vectors, and each interest vector is used for representing one-dimensional interest.
Optionally, the second obtaining module 204 may specifically include:
the obtaining unit is used for obtaining the recommendation score of each candidate video by using a recommendation function based on the interest distribution vector and the prediction index parameter;
the recommendation function comprises a first equalization parameter, wherein the first equalization parameter is used for equalizing the multidimensional interest and the prediction index parameter.
The prediction index parameters comprise click rate information and play completion rate information corresponding to each dimension of interest;
the recommendation function further comprises a second equalization parameter, wherein the second equalization parameter is used for adjusting the relative weight of the click rate information and the play completion rate information.
Optionally, the second model is trained by:
inputting a first training sample to the second model to obtain output of the second model, wherein the first training sample comprises a historical video set sample, the multidimensional interest vector matrix and click rate labels and play completion rate labels of each video in the historical video set sample;
based on a Bayesian personalized ordering algorithm, acquiring a first loss value of the output of the second model;
and updating model parameters of the second model based on the first loss value.
Optionally, the first model is trained by:
inputting a second training sample to the first model to obtain an output of the first model, the second training sample comprising a first time, a video sequence sample, and an interest distribution vector sample;
determining a second loss value of the output of the first model based on the interest distribution vector samples;
and updating model parameters of the first model based on the second loss value.
Optionally, the video training sample includes a target video corresponding to the first time;
the interest distribution vector samples are determined based on similarity of the target video to the multi-dimensional interest vector matrix.
The video recommendation device 200 can implement the processes of the method embodiment of fig. 1 in the embodiment of the present application, and achieve the same beneficial effects, and in order to avoid repetition, the description is omitted here.
The embodiment of the application also provides electronic equipment. Because the principle of solving the problem of the electronic device is similar to that of the video recommendation method shown in fig. 1 in the embodiment of the present application, the implementation of the electronic device may refer to the implementation of the method, and the repetition is not repeated. As shown in fig. 3, the electronic device of the embodiment of the present application includes a memory 320, a transceiver 310, and a processor 300;
a memory 320 for storing a computer program; a transceiver 310 for transceiving data under the control of the processor 300; a processor 300 for reading the computer program in the memory 320 and performing the following operations:
acquiring a video sequence, wherein the video sequence comprises a plurality of videos and playing time of each video;
inputting target time and the video sequence into a first model to predict an interest distribution vector so as to obtain the interest distribution vector corresponding to the target time, wherein the interest distribution vector is used for representing weight distribution of multidimensional interests;
inputting a history video set, a plurality of candidate videos and a multidimensional interest vector matrix which are acquired in advance into a second model to predict index parameters so as to acquire the predicted index parameters of each candidate video;
acquiring recommendation scores of each candidate video based on the interest distribution vector and the predictor parameters;
and recommending the plurality of candidate videos according to the recommendation score of each candidate video.
Wherein in fig. 3, a bus architecture may comprise any number of interconnected buses and bridges, and in particular, one or more processors represented by processor 300 and various circuits of memory represented by memory 320, linked together. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, power management circuits, etc., which are well known in the art and, therefore, will not be described further herein. The bus interface provides an interface. Transceiver 310 may be a number of elements, including a transmitter and a transceiver, providing a means for communicating with various other apparatus over a transmission medium. The processor 300 is responsible for managing the bus architecture and general processing, and the memory 320 may store data used by the processor 300 in performing operations.
The processor 300 may be a central processing unit (Central Processing Unit, CPU), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a Field programmable gate array (Field-Programmable Gate Array, FPGA), or a complex programmable logic device (Complex Programmable Logic Device, CPLD), or may employ a multi-core architecture.
Optionally, the processor 300 is further configured to read the program in the memory 320, and perform the following steps:
acquiring the historical video set;
and extracting features of the historical video set by using a third model to obtain the multidimensional interest vector matrix, wherein the multidimensional interest vector matrix comprises a plurality of interest vectors, and each interest vector is used for representing one-dimensional interest.
Optionally, the obtaining the recommendation score of each candidate video based on the interest distribution vector and the prediction parameter information includes:
acquiring recommendation scores of each candidate video by using a recommendation function based on the interest distribution vector and the predictor parameters;
the recommendation function comprises a first equalization parameter, wherein the first equalization parameter is used for equalizing the multidimensional interest and the prediction index parameter.
Optionally, the predictor parameters include click rate information and play completion rate information corresponding to each dimension of interest;
the recommendation function further comprises a second equalization parameter, wherein the second equalization parameter is used for adjusting the relative weight of the click rate information and the play completion rate information.
Optionally, the second model is trained by:
inputting a first training sample to the second model to obtain output of the second model, wherein the first training sample comprises a historical video set sample, the multidimensional interest vector matrix and click rate labels and play completion rate labels of each video in the historical video set sample;
based on a Bayesian personalized ordering algorithm, acquiring a first loss value of the output of the second model;
and updating model parameters of the second model based on the first loss value.
Optionally, the first model is trained by:
inputting a second training sample to the first model to obtain an output of the first model, the second training sample comprising a first time, a video sequence sample, and an interest distribution vector sample;
determining a second loss value of the output of the first model based on the interest distribution vector samples;
and updating model parameters of the first model based on the second loss value.
Optionally, the video training sample includes a target video corresponding to the first time;
the interest distribution vector samples are determined based on similarity of the target video to the multi-dimensional interest vector matrix.
The electronic device provided in the embodiment of the present application may execute the method embodiment shown in fig. 1, and its implementation principle and technical effects are similar, and this embodiment is not repeated here.
The present disclosure further provides a readable storage medium, where a program is stored, where the program when executed by a processor implements the processes of the method embodiment shown in fig. 1, and the same technical effects can be achieved, and for avoiding repetition, a detailed description is omitted herein.
In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
In addition, each functional unit in each embodiment of the present application may be integrated in one processing unit, or each unit may be physically included separately, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in hardware plus software functional units.
The integrated units implemented in the form of software functional units described above may be stored in a computer readable storage medium. The software functional unit is stored in a storage medium, and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform part of the steps of the transceiving method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
While the foregoing is directed to the preferred embodiments of the present application, it should be noted that modifications and adaptations to those embodiments may occur to one skilled in the art and that such modifications and adaptations are intended to be comprehended within the scope of the present application without departing from the principles set forth herein.
Claims (10)
1. A video recommendation method, comprising:
acquiring a video sequence, wherein the video sequence comprises a plurality of videos and playing time of each video;
inputting target time and the video sequence into a first model to predict an interest distribution vector so as to obtain the interest distribution vector corresponding to the target time, wherein the interest distribution vector is used for representing weight distribution of multidimensional interests;
inputting a history video set, a plurality of candidate videos and a multidimensional interest vector matrix which are acquired in advance into a second model to predict index parameters so as to acquire the predicted index parameters of each candidate video;
acquiring recommendation scores of each candidate video based on the interest distribution vector and the predictor parameters;
and recommending the plurality of candidate videos according to the recommendation score of each candidate video.
2. The method of claim 1, wherein before the inputting the target time and the video sequence into the first model for predicting the interest distribution vector, the method further comprises:
acquiring the historical video set;
and extracting features of the historical video set by using a third model to obtain the multidimensional interest vector matrix, wherein the multidimensional interest vector matrix comprises a plurality of interest vectors, and each interest vector is used for representing one-dimensional interest.
3. The method of claim 2, wherein the obtaining a recommendation score for each candidate video based on the interest distribution vector and the prediction parameter information comprises:
acquiring recommendation scores of each candidate video by using a recommendation function based on the interest distribution vector and the predictor parameters;
the recommendation function comprises a first equalization parameter, wherein the first equalization parameter is used for equalizing the multidimensional interest and the prediction index parameter.
4. The method of claim 3, wherein the predictor parameters include click-through rate information and completion rate information corresponding to each dimension of interest;
the recommendation function further comprises a second equalization parameter, wherein the second equalization parameter is used for adjusting the relative weight of the click rate information and the play completion rate information.
5. The method of claim 4, wherein the second model is trained by:
inputting a first training sample to the second model to obtain output of the second model, wherein the first training sample comprises a historical video set sample, the multidimensional interest vector matrix and click rate labels and play completion rate labels of each video in the historical video set sample;
based on a Bayesian personalized ordering algorithm, acquiring a first loss value of the output of the second model;
and updating model parameters of the second model based on the first loss value.
6. The method according to any one of claims 1 to 5, wherein the first model is trained by:
inputting a second training sample to the first model to obtain an output of the first model, the second training sample comprising a first time, a video sequence sample, and an interest distribution vector sample;
determining a second loss value of the output of the first model based on the interest distribution vector samples;
and updating model parameters of the first model based on the second loss value.
7. The method of claim 6, wherein the video training sample includes a target video corresponding to the first time;
the interest distribution vector samples are determined based on similarity of the target video to the multi-dimensional interest vector matrix.
8. A video recommendation device, comprising:
the first acquisition module is used for acquiring a video sequence, wherein the video sequence comprises a plurality of videos and the playing time of each video;
the first prediction module is used for inputting the target time and the video sequence into a first model to predict an interest distribution vector so as to obtain the interest distribution vector corresponding to the target time, wherein the interest distribution vector is used for representing weight distribution of multidimensional interests;
the second prediction module is used for inputting a history video set, a plurality of candidate videos and a multidimensional interest vector matrix which are acquired in advance into a second model to predict index parameters so as to acquire the prediction index parameters of each candidate video;
the second acquisition module is used for acquiring the recommendation score of each candidate video based on the interest distribution vector and the prediction index parameter;
and the recommending module is used for recommending the plurality of candidate videos according to the recommending score of each candidate video.
9. An electronic device, comprising: a transceiver, a memory, a processor, and a computer program stored on the memory and executable on the processor; it is characterized in that the method comprises the steps of,
the processor being configured to read a program in a memory to implement the steps in the method according to any one of claims 1 to 7.
10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111613794.5A CN116366923A (en) | 2021-12-27 | 2021-12-27 | Video recommendation method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111613794.5A CN116366923A (en) | 2021-12-27 | 2021-12-27 | Video recommendation method and device and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116366923A true CN116366923A (en) | 2023-06-30 |
Family
ID=86925228
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111613794.5A Pending CN116366923A (en) | 2021-12-27 | 2021-12-27 | Video recommendation method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116366923A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117648462A (en) * | 2024-01-29 | 2024-03-05 | 深圳感臻智能股份有限公司 | Video recommendation method and system |
-
2021
- 2021-12-27 CN CN202111613794.5A patent/CN116366923A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117648462A (en) * | 2024-01-29 | 2024-03-05 | 深圳感臻智能股份有限公司 | Video recommendation method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110941740B (en) | Video recommendation method and computer-readable storage medium | |
| US20210326674A1 (en) | Content recommendation method and apparatus, device, and storage medium | |
| CN107404656B (en) | Live video recommended method, device and server | |
| US9430532B2 (en) | Media content rankings for discovery of novel content | |
| CN108595493B (en) | Media content pushing method and device, storage medium and electronic device | |
| WO2022228303A1 (en) | Video processing method, and storage medium and processor | |
| CN112597395A (en) | Object recommendation method, device, equipment and storage medium | |
| CN112507163B (en) | Duration prediction model training method, recommendation method, device, equipment and medium | |
| CN111858969B (en) | Multimedia data recommendation method, device, computer equipment and storage medium | |
| CN113535991B (en) | Multimedia resource recommendation method and device, electronic equipment and storage medium | |
| CN113742567B (en) | Recommendation method and device for multimedia resources, electronic equipment and storage medium | |
| CN111597446B (en) | Content pushing method and device based on artificial intelligence, server and storage medium | |
| CN112100221B (en) | Information recommendation method and device, recommendation server and storage medium | |
| CN112579913A (en) | Video recommendation method, device, equipment and computer-readable storage medium | |
| Aharon et al. | Watch-it-next: a contextual TV recommendation system | |
| CN111400609A (en) | User recommendation method and device, storage medium and server | |
| US8688716B1 (en) | Recommending pairwise video comparisons to improve ranking | |
| US20150186934A1 (en) | Group composition based recommender system and method | |
| CN113919923B (en) | Live broadcast recommendation model training method, live broadcast recommendation method and related equipment | |
| CN116366923A (en) | Video recommendation method and device and electronic equipment | |
| CN112269943B (en) | Information recommendation system and method | |
| CN113836388A (en) | Information recommendation method and device, server and storage medium | |
| CN115618121B (en) | Personalized information recommendation method, device, equipment and storage medium | |
| CN106815285A (en) | The method of the video recommendations based on video website, device and electronic equipment | |
| CN110418171B (en) | Media resource pushing method and device, storage medium and electronic device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |