CN105975641A - Video recommendation method ad device - Google Patents
Video recommendation method ad device Download PDFInfo
- Publication number
- CN105975641A CN105975641A CN201610561318.6A CN201610561318A CN105975641A CN 105975641 A CN105975641 A CN 105975641A CN 201610561318 A CN201610561318 A CN 201610561318A CN 105975641 A CN105975641 A CN 105975641A
- Authority
- CN
- China
- Prior art keywords
- video
- videos
- list
- targeted customer
- represent
- 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
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a video recommendation method and device. The method includes the steps that user behavior data of a target user are acquired, and a first video list is generated according to the user behavior data; user characteristic information of the target user is obtained, and a user characteristic vector is generated according to the user characteristic information; a prediction model is determined according to the first video list and the user characteristic vector; prediction values of all videos to be selected are calculated according to the prediction model; the videos to be recommended are selected from all the videos to be selected according to the prediction values of the videos to be selected, and a video recommendation list is generated according to the videos to be recommended. By means of the video recommendation method and device, video recommendation can be conducted in combination with multiple videos related to the user, the user characteristic vector acts on the prediction model, and thus accuracy of video recommendation can be improved.
Description
Technical field
The present invention relates to areas of information technology, particularly relate to a kind of video recommendation method and device.
Background technology
Internet era be epoch of information explosion, the quantity of information increases with exponential.At video field, the number of video
Amount increases with blowout formula.At present, the YouTube website video of about 60 hours per minute is uploaded, and YouTube website is total
Number of videos reached several hundred million.User is before viewing video, it is often necessary to take a significant amount of time and it just can be found interested
Video, Consumer's Experience is poor.
Video personalized recommendation technology can browsing and viewing behavior according to user, recommend it may be interested to user
Video.The most conventional personalized recommendation algorithm includes: proposed algorithm based on correlation rule, content-based recommendation algorithm
With proposed algorithm based on collaborative filtering.In the personalized recommendation of video website, personalized recommendation based on collaborative filtering is calculated
Method is the most commonly used.
During realizing the present invention, inventor finds that prior art at least there is problems in that existing video pushes away
The technology of recommending only considered user's viewing behavior for single video, and the accuracy causing video recommendations is relatively low.
Summary of the invention
Technical problem
In view of this, the technical problem to be solved in the present invention is, existing video recommendations technology only considered user for
The viewing behavior of single video, the accuracy causing video recommendations is relatively low.
Solution
In order to solve above-mentioned technical problem, according to one embodiment of the invention, it is provided that a kind of video recommendation method, bag
Include:
Gather the user behavior data of targeted customer, and generate the first list of videos according to described user behavior data；
Obtain the user's characteristic information of described targeted customer, and according to described user's characteristic information generate user characteristics to
Amount；
Forecast model is determined according to described first list of videos and described user characteristics vector；
The predictive value of all videos to be selected is calculated according to described forecast model；
Predictive value according to described video to be selected filters out video to be recommended from all described videos to be selected, and according to institute
State video to be recommended and generate video recommendations list.
For said method, in a kind of possible implementation, gather the user behavior data of targeted customer, according to institute
State user behavior data and generate the first list of videos, including:
All user behavior datas of the described targeted customer in the time period are specified in collection；
Effective user behavior data is filtered out from the user behavior data gathered；
Described effective user behavior data is ranked up by the time corresponding according to described effective user behavior data,
Obtain described first list of videos.
For said method, in a kind of possible implementation, special according to described first list of videos and described user
Levy vector and determine forecast model, including:
Supervision vector as shown in Equation 1 is determined according to described user characteristics vector；
Wherein, uiRepresent i-th targeted customer, E (ui) represent that the supervision of described i-th targeted customer is vectorial, H (ui)kTable
Show the kth user characteristics vector of described i-th targeted customer, MkRepresent the kth user characteristics of described i-th targeted customer
The weight of vector；
Use shot and long term memory recurrent neural network, determine according to described first list of videos and described supervision vector described
Forecast model.
For said method, in a kind of possible implementation, calculate all videos to be selected according to described forecast model
Predictive value, particularly as follows:
Employing formula 2 calculates the predictive value of described video to be selected respectively；
Wherein, vjRepresent jth video to be selected, vtRepresent the t video in described first list of videos, uiRepresent i-th
Individual targeted customer, p (vj|vt,ui) represent the predictive value of described jth video to be selected, s (vt,ui)jWith s (vt,ui)mAccording to formula 3
Or formula 4 determines；
Wherein, l represents the l neuron of the output layer of neutral net, v1Represent the 1st in described first list of videos
Individual video, t is more than 1, and n represents the n-th neuron of the hidden layer of described neutral net, lstm (v1,ui)n、lstm(vt-1,ui)
With lstm (vt,ui,lstm(vt-1,ui))nDetermine according to described forecast model, wlnRepresent described neutral net hidden layer
N neuron is for the weight of l neuron of the output layer of described neutral net.
For said method, in a kind of possible implementation, according to the predictive value of described video to be selected from all institutes
State and video to be selected filters out video to be recommended, including:
From all described videos to be selected, filter out predictive value more than the video described to be selected setting threshold value, obtain second and regard
Frequently list；
According to following at least one from described second list of videos, filter out described video to be recommended: described video to be selected
Uploader information, the channel information belonging to described video to be selected, described targeted customer watch the data of video and described target
The interest tags of user.
In order to solve above-mentioned technical problem, according to another embodiment of the present invention, it is provided that a kind of video recommendations device, bag
Include:
First list of videos generation module, for gathering the user behavior data of targeted customer, and according to described user's row
For data genaration the first list of videos；
User characteristics vector generation module, for obtaining the user's characteristic information of described targeted customer, and according to described use
Family characteristic information generates user characteristics vector；
Forecast model determines module, for determining prediction mould according to described first list of videos and described user characteristics vector
Type；
Predictor calculation module, for calculating the predictive value of all videos to be selected according to described forecast model；
Video recommendations List Generating Module, is used for the predictive value according to described video to be selected from all described videos to be selected
Filter out video to be recommended, and generate video recommendations list according to described video to be recommended.
For said apparatus, in a kind of possible implementation, described first list of videos generation module includes:
User behavior data gathers submodule, all user's row of the described targeted customer in gathering the appointment time period
For data；
User behavior data screening submodule, for filtering out effective user's row from the user behavior data gathered
For data；
Sorting sub-module, for the time corresponding according to described effective user behavior data to described effective user's row
It is ranked up for data, obtains described first list of videos.
For said apparatus, in a kind of possible implementation, described forecast model determines that module includes:
Supervision vector determines submodule, for determining supervision vector as shown in Equation 1 according to described user characteristics vector；
Wherein, uiRepresent i-th targeted customer, E (ui) represent that the supervision of described i-th targeted customer is vectorial, H (ui)kTable
Show the kth user characteristics vector of described i-th targeted customer, MkRepresent the kth user characteristics of described i-th targeted customer
The weight of vector；
Forecast model determines submodule, is used for using shot and long term to remember recurrent neural network, according to described first video row
Table and described supervision vector determine described forecast model.
For said apparatus, in a kind of possible implementation, described predictor calculation module specifically for:
Employing formula 2 calculates the predictive value of described video to be selected respectively；
Wherein, vjRepresent jth video to be selected, vtRepresent the t video in described first list of videos, uiRepresent i-th
Individual targeted customer, p (vj|vt,ui) represent the predictive value of described jth video to be selected, s (vt,ui)jWith s (vt,ui)mAccording to formula 3
Or formula 4 determines；
Wherein, l represents the l neuron of the output layer of neutral net, v1Represent the 1st in described first list of videos
Individual video, t is more than 1, and n represents the n-th neuron of the hidden layer of described neutral net, lstm (v1,ui)n、lstm(vt-1,ui)
With lstm (vt,ui,lstm(vt-1,ui))nDetermine according to described forecast model, wlnRepresent described neutral net hidden layer
N neuron is for the weight of l neuron of the output layer of described neutral net.
For said apparatus, in a kind of possible implementation, described video recommendations List Generating Module includes:
First screening submodule, for filtering out predictive value more than setting described in threshold value from all described videos to be selected
Video to be selected, obtains the second list of videos；
Second screening submodule, for according to following at least one filter out from described second list of videos described in wait to push away
Recommend video: the channel information belonging to the uploader information of described video to be selected, described video to be selected, described targeted customer viewing regard
The data of frequency and the interest tags of described targeted customer.
Beneficial effect
Obtain the first list of videos by gathering the user behavior data of targeted customer, generate according to user's characteristic information and use
Family characteristic vector, determines forecast model in conjunction with the first list of videos and user characteristics vector, carries out video further according to forecast model
Recommending, video recommendation method and device according to embodiments of the present invention can carry out video in conjunction with user-dependent multiple videos
Recommend, and user characteristics vector is acted on forecast model such that it is able to improve the accuracy of video recommendations.
According to below with reference to the accompanying drawings detailed description of illustrative embodiments, the further feature of the present invention and aspect being become
Clear.
Accompanying drawing explanation
The accompanying drawing of the part comprising in the description and constituting description together illustrates the present invention's with description
Exemplary embodiment, feature and aspect, and for explaining the principle of the present invention.
Fig. 1 illustrates the flowchart of video recommendation method according to an embodiment of the invention；
Fig. 2 illustrates that the one of video recommendation method step S101 according to an embodiment of the invention exemplary implements stream
Cheng Tu；
It is pre-that Fig. 3 illustrates in video recommendation method step S105 according to an embodiment of the invention according to described video to be selected
What measured value filtered out video to be recommended from all described videos to be selected one exemplary implements flow chart；
Fig. 4 illustrates the structured flowchart of video recommendations device according to another embodiment of the present invention；
Fig. 5 illustrates an exemplary structured flowchart of video recommendations device according to another embodiment of the present invention；
Fig. 6 shows the structured flowchart of a kind of video recommendations equipment of an alternative embodiment of the invention.
Detailed description of the invention
Various exemplary embodiments, feature and the aspect of the present invention is described in detail below with reference to accompanying drawing.In accompanying drawing identical
Reference represent the same or analogous element of function.Although the various aspects of embodiment shown in the drawings, but remove
Non-specifically is pointed out, it is not necessary to accompanying drawing drawn to scale.
The most special word " exemplary " means " as example, embodiment or illustrative ".Here as " exemplary "
Illustrated any embodiment should not necessarily be construed as preferred or advantageous over other embodiments.
It addition, in order to better illustrate the present invention, detailed description of the invention below gives numerous details.
It will be appreciated by those skilled in the art that do not have some detail, the present invention equally implements.In some instances, for
Method well known to those skilled in the art, means, element and circuit are not described in detail, in order to highlight the purport of the present invention.
Embodiment 1
Fig. 1 illustrates the flowchart of video recommendation method according to an embodiment of the invention.As it is shown in figure 1, the method
Specifically include that
In step S101, gather the user behavior data of targeted customer, and regard according to user behavior data generation first
Frequently list.
Wherein, the user behavior data of targeted customer can include but not limited to following at least one: targeted customer watch
The data of video, targeted customer comment on the data of video, targeted customer subscribes to the data of video and video is stepped on targeted customer top
Data.
In step s 102, obtain the user's characteristic information of targeted customer, and it is special to generate user according to user's characteristic information
Levy vector.
As an example of the embodiment of the present invention, the user's characteristic information obtaining targeted customer can be: from daily record literary composition
Part obtains the user's characteristic information of targeted customer.
Wherein, the user's characteristic information of targeted customer can include but not limited to following at least one: targeted customer's is emerging
Interest label, IP (Internet Protocol, the Internet protocol) address of targeted customer, the facility information of targeted customer, target
Number of videos, the age of targeted customer and the sex of targeted customer of user's viewing.Wherein, the IP address of targeted customer can be used
In the geographical position of record object user, thus it can be considered that the region of targeted customer in video recommendations.Targeted customer's
Facility information can include following at least one: the device type of targeted customer, the unit type of targeted customer and targeted customer
The operating system that used of equipment.Such as, the device type of targeted customer can be PC (Personal Computer, individual
Computer), panel computer or mobile phone etc.；The operating system that the equipment of targeted customer is used can be Android or iOS
Deng.
As an example of the embodiment of the present invention, when determining the interest tags of targeted customer, it may be considered that target is used
The time scale of each video is watched at family.Such as, if targeted customer watches the time span of a certain video when accounting for this video total
Between the ratio of length less, then when determining the interest tags of targeted customer, the marking value to this video can be reduced；If target
User watch the time span of a certain video account for this video total time length large percentage, or targeted customer watches this and regards
The time span of frequency more than the total time length of this video, then, when determining the interest tags of targeted customer, can improve this
The marking value of video.
As an example of the embodiment of the present invention, the number of videos of targeted customer's viewing can pass through demarcation interval and two
The method of value processes.For example, it is possible to number of videos to be divided into first interval, the second interval and the 3rd interval.Firstth district
Between number of videos can be 0 to 5, the number of videos in the second interval can be 6 to 34, and the number of videos in the 3rd interval can be
35 to 100.The mark in the first interval can be (0,0), and the mark in the second interval can be (0,1), and the mark in the 3rd interval can
Think (1,0).If the number of videos of targeted customer's viewing is 90, then interval corresponding to the 3rd.
Such as, the user's characteristic information of targeted customer includes the interest tags of targeted customer, the IP address of targeted customer, mesh
The facility information of mark user and the number of videos of targeted customer's viewing, as an example of the embodiment of the present invention, according to user
Characteristic information generates user characteristics vector: represents the interest tags of targeted customer with FT position, represents that target is used with FI position
The IP address at family, represents the facility information of targeted customer with FP position, represents, with FV position, the number of videos that targeted customer watches.Example
As, if FT position is (0,0), FI position is (1,0), and FP position is (1,1), and FV position is (1,0), then user characteristics vector can be expressed as
(0,0,1,0,1,1,1,0)。
In step s 103, forecast model is determined according to the first list of videos and user characteristics vector.
As an example of the embodiment of the present invention, Recognition with Recurrent Neural Network (Recurrent Neural can be used
Networks, RNNs) determine forecast model.This Recognition with Recurrent Neural Network can include input layer, hidden layer and output layer.Its
In, the input of input layer can be the first list of videos；Hidden layer can use shot and long term to remember (Long-Short Term
Memory, LSTM) recurrent neural network realizes, the problem that during to eliminate gradient updating, gradient disappears, can obtain more simultaneously
Good forecast model；The output of output layer can be forecast model.In this example, it is also possible to user characteristics vector is introduced and follows
In ring neutral net, to play the effect of monitoring forecast model.
It should be noted that the shot and long term memory recurrent neural network in this example can utilize existing shot and long term to remember
Recursive neural network technology realizes, and no longer repeats the operation principle of shot and long term memory recurrent neural network at this.
In step S104, calculate the predictive value of all videos to be selected according to forecast model.
In step S105, from all videos to be selected, filter out video to be recommended according to the predictive value of video to be selected, and
Video recommendations list is generated according to video to be recommended.
After step S105, the method can also include: video recommendations list is recommended targeted customer.
Fig. 2 illustrates that the one of video recommendation method step S101 according to an embodiment of the invention exemplary implements stream
Cheng Tu.As in figure 2 it is shown, gather the user behavior data of targeted customer, generate the first list of videos according to user behavior data, bag
Include:
In step s 201, all user behavior datas of the targeted customer in the time period is specified in collection.
In step S202, from the user behavior data gathered, filter out effective user behavior data.
For example, it is possible to the user behavior data repeating to watch video is defined as invalid user behavior data, it is also possible to
User behavior data the least for the completed percentage of viewing video is defined as invalid user behavior data, in this no limit.
In step S203, effective user behavior data is carried out by the time corresponding according to effective user behavior data
Sequence, obtains the first list of videos.
Wherein, when the time that effective user behavior data is corresponding can be the generation of this effective user behavior data
Between.According to time corresponding to effective user behavior data effective user behavior data is ranked up can be: according to having
Effective user behavior data is ranked up by the user behavior data of effect time sequencing from the near to the remote.
It should be noted that targeted customer is after viewing the first video, the second video and the 3rd video, with targeted customer
After viewing the second video, the 3rd video and the first video, it is desirable to the video of viewing is probably different.This example considers
The time of origin of effective user behavior data such that it is able to improve the accuracy rate of video recommendations.
As an example of the embodiment of the present invention, after obtaining the first list of videos, this first video can be stored
List.
In a kind of possible implementation, determine forecast model according to the first list of videos and user characteristics vector, bag
Include: determine supervision vector as shown in Equation 1 according to user characteristics vector；
Wherein, uiRepresent i-th targeted customer, E (ui) represent that the supervision of i-th targeted customer is vectorial, H (ui)kRepresent the
The kth user characteristics vector of i targeted customer, MkRepresent the weight of the kth user characteristics vector of i-th targeted customer；
Use shot and long term memory recurrent neural network, determine forecast model according to the first list of videos and supervision vector.
As an example of the embodiment of the present invention, determine forecast model according to the first list of videos and user characteristics vector
May include that and determine supervision vector according to user characteristics vector, by supervision vector monitoring forecast model, thus to improve and to regard
The precision that frequency is recommended.
As an example of the embodiment of the present invention, before using shot and long term memory recurrent neural network, the method is also
May include that and the video in the first list of videos is mapped as video vector, and the value of video vector mapping obtained controls
For more than 0 and less than 1.Here map the video vector obtained and training later can update weight, thus reach video
The training of vector.
In a kind of possible implementation, calculate the predictive value of all videos to be selected according to forecast model, particularly as follows: adopt
The predictive value of video to be selected is calculated respectively by formula 2；
Wherein, vjRepresent jth video to be selected, vtRepresent the t video in the first list of videos, uiRepresent i-th mesh
Mark user, p (vj|vt,ui) represent jth video to be selected predictive value, s (vt,ui)jWith s (vt,ui)mTrue according to formula 3 or formula 4
Fixed；
Wherein, l represents the l neuron of the output layer of neutral net, v1Represent that the in the first list of videos the 1st regards
Frequently, t is more than 1, and n represents the n-th neuron of the hidden layer of neutral net, lstm (v1,ui)n、lstm(vt-1,ui) and lstm
(vt,ui,lstm(vt-1,ui))nDetermine according to forecast model, wlnRepresent neutral net hidden layer the n-th neuron for
The weight of l neuron of the output layer of neutral net.
It should be noted that in this example, for the output lstm (v of first hidden layer1,ui)n, employing formula 3 is carried out
Softmax converts；From the beginning of second hidden layer, employing formula 4 carries out softmax conversion, to consider the defeated of a upper hidden layer
Go out information.In this example, lstm represents the processing procedure of shot and long term memory recurrent neural network, shot and long term memory recurrent neural
The processing procedure of network can utilize existing shot and long term memory recursive neural network technology to realize, and no longer remembers shot and long term at this
The processing procedure recalling recurrent neural network repeats.
Fig. 3 illustrates the predictive value in video recommendation method step S105 according to an embodiment of the invention according to video to be selected
Filter out video to be recommended from all videos to be selected one exemplary implements flow chart.As it is shown on figure 3, according to be selected
The predictive value of video filters out video to be recommended from all videos to be selected, including:
In step S301, from all videos to be selected, filter out predictive value more than the video to be selected setting threshold value, obtain
Second list of videos.
In step s 302, according to following at least one from the second list of videos, filter out video to be recommended: to be selected regard
Channel information, targeted customer belonging to the uploader information of frequency, video to be selected watch the data of video and the interest of targeted customer
Label.
As an example of the embodiment of the present invention, filter out from all videos to be selected according to the predictive value of video to be selected
Video to be recommended can include the video to be selected filtering out predictive value from all videos to be selected more than setting threshold value, obtains second
List of videos, it is thus possible to the video that prediction user may be interested.In this example, always according to following at least one regard from second
List filters out video to be recommended frequently, so that the video recommended possesses multiformity: the uploader information of video to be selected, to be selected regard
Channel information belonging to Pin, targeted customer watch the data of video and the interest tags of targeted customer.Such as, if the second video arranges
Table includes the video that more than four same uploader are uploaded, then can retain predictive value ranking in the video that this uploader is uploaded
The video of first three is as video to be recommended.The most such as, if the second list of videos includes regarding of more than four same two grades of channels
Frequently, then first three video of predictive value ranking in the video of these two grades of channels can be retained as video to be recommended.Such as, variety frequency
Road is a certain level channel, and Hunan XATV-6 is two grades of channels under this level channel.The most such as, if in the second list of videos
Including the video under more than four same three grades of interest tags, then can retain predictive value in the video under these three grades of interest tags
First three video of ranking is as video to be recommended.Such as, one-level interest tags is amusement, and star in amusement circle is this one-level interest tags
Under two grades of interest tags, Beyond is three grades of interest tags under these two grades of interest tags.The most such as, if the second list of videos
Include the video that targeted customer watched in the recent period, the most not using this video as video to be recommended.
As an example of the embodiment of the present invention, after step S302, push away generating video according to video to be recommended
Before recommending list, the method can also include: if the quantity of video to be recommended is more than Q, then from according to predictive value from big to small
Video to be recommended is ranked up by order.Generating video recommendations list according to video to be recommended can be: to be recommended according to front Q
Video generates video recommendations list.Video hotspot list can also be generated according to other video to be recommended, then by video recommendations
Targeted customer is recommended in list and video hotspot list.
So, the first list of videos is obtained, according to user's characteristic information by the user behavior data of collection targeted customer
Generation user characteristics vector, determines forecast model in conjunction with the first list of videos and user characteristics vector, enters further according to forecast model
Row video recommendations, video recommendation method according to embodiments of the present invention can carry out video in conjunction with user-dependent multiple videos
Recommend, and user characteristics vector is acted on forecast model such that it is able to improve the accuracy of video recommendations.
Embodiment 2
Fig. 4 illustrates the structured flowchart of video recommendations device according to another embodiment of the present invention.This device may be used for fortune
Video recommendation method shown in row Fig. 1 to Fig. 3.For convenience of description, illustrate only the part relevant to the embodiment of the present invention.
As shown in Figure 4, this device includes: the first list of videos generation module 41, for gathering user's row of targeted customer
For data, and generate the first list of videos according to described user behavior data；User characteristics vector generation module 42, is used for obtaining
The user's characteristic information of described targeted customer, and generate user characteristics vector according to described user's characteristic information；Forecast model is true
Cover half block 43, for determining forecast model according to described first list of videos and described user characteristics vector；Predictor calculation mould
Block 44, for calculating the predictive value of all videos to be selected according to described forecast model；Video recommendations List Generating Module 45, is used for
Predictive value according to described video to be selected filters out video to be recommended from all described videos to be selected, and according to described to be recommended
Video generates video recommendations list.
Fig. 5 illustrates an exemplary structured flowchart of video recommendations device according to another embodiment of the present invention.Fig. 5 gets the bid
Number assembly identical with Fig. 4 has identical function, for simplicity's sake, omits the detailed description to these assemblies.Such as Fig. 5 institute
Show:
In a kind of possible implementation, described first list of videos generation module includes: user behavior data collection
Submodule 411, all user behavior datas of the described targeted customer in gathering the appointment time period；User behavior data sieves
Select submodule 412, for filtering out effective user behavior data from the user behavior data gathered；Sorting sub-module
413, for described effective user behavior data being ranked up according to the time that described effective user behavior data is corresponding,
Obtain described first list of videos.
In a kind of possible implementation, described forecast model determines that module 43 includes:
Supervision vector determines submodule 431, for according to described user characteristics vector determine supervision as shown in Equation 1 to
Amount；
Wherein, uiRepresent i-th targeted customer, E (ui) represent that the supervision of described i-th targeted customer is vectorial, H (ui)kTable
Show the kth user characteristics vector of described i-th targeted customer, MkRepresent the kth user characteristics of described i-th targeted customer
The weight of vector；
Forecast model determines submodule 432, is used for using shot and long term to remember recurrent neural network, according to described first video
List and described supervision vector determine described forecast model.
In a kind of possible implementation, described predictor calculation module 44 specifically for:
Employing formula 2 calculates the predictive value of described video to be selected respectively；
Wherein, vjRepresent jth video to be selected, vtRepresent the t video in described first list of videos, uiRepresent i-th
Individual targeted customer, p (vj|vt,ui) represent the predictive value of described jth video to be selected, s (vt,ui)jWith s (vt,ui)mAccording to formula 3
Or formula 4 determines；
Wherein, l represents the l neuron of the output layer of neutral net, v1Represent the 1st in described first list of videos
Individual video, t is more than 1, and n represents the n-th neuron of the hidden layer of described neutral net, lstm (v1,ui)n、lstm(vt-1,ui)
With lstm (vt,ui,lstm(vt-1,ui))nDetermine according to described forecast model, wlnRepresent described neutral net hidden layer
N neuron is for the weight of l neuron of the output layer of described neutral net.
In a kind of possible implementation, described video recommendations List Generating Module 45 includes: the first screening submodule
451, for filtering out predictive value from all described videos to be selected more than the video described to be selected setting threshold value, obtain second and regard
Frequently list；Second screening submodule 452, for according to following at least one filter out from described second list of videos described in treat
Recommend video: the channel information belonging to the uploader information of described video to be selected, described video to be selected, described targeted customer watch
The data of video and the interest tags of described targeted customer.
It should be noted that so, obtain the first list of videos by gathering the user behavior data of targeted customer, according to
User's characteristic information generates user characteristics vector, determines forecast model, then root in conjunction with the first list of videos and user characteristics vector
It is predicted that model carries out video recommendations, video recommendations device according to embodiments of the present invention can be in conjunction with multiple with user-dependent
Video carries out video recommendations, and user characteristics vector is acted on forecast model such that it is able to improve the accuracy of video recommendations.
Embodiment 3
Fig. 6 shows the structured flowchart of a kind of video recommendations equipment of an alternative embodiment of the invention.Described video pushes away
Recommending equipment 1100 can be to possess the host server of computing capability, personal computer PC or portable portable computing
Machine or terminal etc..Calculating node is not implemented and limits by the specific embodiment of the invention.
Described video recommendations equipment 1100 includes processor (processor) 1110, communication interface (Communications
Interface) 1120, memorizer (memory) 1130 and bus 1140.Wherein, processor 1110, communication interface 1120 and
Memorizer 1130 completes mutual communication by bus 1140.
Communication interface 1120 is used for and network device communications, and wherein the network equipment includes such as Virtual Machine Manager center, is total to
Enjoy storage etc..
Processor 1110 is used for performing program.Processor 1110 is probably a central processor CPU, or special collection
Become circuit ASIC (Application Specific Integrated Circuit), or be configured to implement the present invention
One or more integrated circuits of embodiment.
Memorizer 1130 is used for depositing file.Memorizer 1130 may comprise high-speed RAM memorizer, it is also possible to also includes non-
Volatile memory (non-volatile memory), for example, at least one disk memory.Memorizer 1130 can also be to deposit
Memory array.Memorizer 1130 is also possible to by piecemeal, and described piece can be by certain rule sets synthesis virtual volume.
In a kind of possible embodiment, said procedure can be the program code including computer-managed instruction.This journey
Sequence is particularly used in: realize the operation of each step in embodiment 1.
Those of ordinary skill in the art are it is to be appreciated that each exemplary cell in embodiment described herein and algorithm
Step, it is possible to being implemented in combination in of electronic hardware or computer software and electronic hardware.These functions are actually with hardware also
It is that software form realizes, depends on application-specific and the design constraint of technical scheme.Professional and technical personnel can be for
Specific application selects different methods to realize described function, but this realization is it is not considered that exceed the model of the present invention
Enclose.
If using the form of computer software realize described function and as independent production marketing or use time, then exist
To a certain extent it is believed that all or part of (part such as contributed prior art) of technical scheme is
Embody in form of a computer software product.This computer software product is generally stored inside the non-volatile of embodied on computer readable
In storage medium, including some instructions with so that computer equipment (can be that personal computer, server or network set
Standby etc.) perform all or part of step of various embodiments of the present invention method.And aforesaid storage medium include USB flash disk, portable hard drive,
Read only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic
The various medium that can store program code such as dish or CD.
The above, the only detailed description of the invention of the present invention, but protection scope of the present invention is not limited thereto, and any
Those familiar with the art, in the technical scope that the invention discloses, can readily occur in change or replace, should contain
Cover within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with described scope of the claims.
Claims (10)
1. a video recommendation method, it is characterised in that including:
Gather the user behavior data of targeted customer, and generate the first list of videos according to described user behavior data；
Obtain the user's characteristic information of described targeted customer, and generate user characteristics vector according to described user's characteristic information；
Forecast model is determined according to described first list of videos and described user characteristics vector；
The predictive value of all videos to be selected is calculated according to described forecast model；
Predictive value according to described video to be selected filters out video to be recommended from all described videos to be selected, and treats according to described
Video is recommended to generate video recommendations list.
Method the most according to claim 1, it is characterised in that gather the user behavior data of targeted customer, according to described
User behavior data generates the first list of videos, including:
All user behavior datas of the described targeted customer in the time period are specified in collection；
Effective user behavior data is filtered out from the user behavior data gathered；
Described effective user behavior data is ranked up by the time corresponding according to described effective user behavior data, obtains
Described first list of videos.
Method the most according to claim 1, it is characterised in that according to described first list of videos and described user characteristics to
Amount determines forecast model, including:
Supervision vector as shown in Equation 1 is determined according to described user characteristics vector；
Wherein, uiRepresent i-th targeted customer, E (ui) represent that the supervision of described i-th targeted customer is vectorial, H (ui)kRepresent institute
State the kth user characteristics vector of i-th targeted customer, MkRepresent the kth user characteristics vector of described i-th targeted customer
Weight；
Use shot and long term memory recurrent neural network, determine described prediction according to described first list of videos and described supervision vector
Model.
4. according to the method described in claims 1 to 3 any one, it is characterised in that calculate all according to described forecast model
The predictive value of video to be selected, particularly as follows:
Employing formula 2 calculates the predictive value of described video to be selected respectively；
Wherein, vjRepresent jth video to be selected, vtRepresent the t video in described first list of videos, uiRepresent i-th mesh
Mark user, p (vj|vt,ui) represent the predictive value of described jth video to be selected, s (vt,ui)jWith s (vt,ui)mAccording to formula 3 or formula 4
Determine；
Wherein, l represents the l neuron of the output layer of neutral net, v1Represent that the in described first list of videos the 1st regards
Frequently, t is more than 1, and n represents the n-th neuron of the hidden layer of described neutral net, lstm (v1,ui)n、lstm(vt-1,ui) and
lstm(vt,ui,lstm(vt-1,ui))nDetermine according to described forecast model, wlnRepresent described neutral net hidden layer n-th
Individual neuron is for the weight of l neuron of the output layer of described neutral net.
5. according to the method described in claims 1 to 3 any one, it is characterised in that according to the predictive value of described video to be selected
Video to be recommended is filtered out from all described videos to be selected, including:
From all described videos to be selected, filter out predictive value more than the video described to be selected setting threshold value, obtain the second video row
Table；
According to following at least one from described second list of videos, filter out described video to be recommended: described video to be selected upper
Channel information, described targeted customer belonging to biography person's information, described video to be selected watch the data of video and described targeted customer
Interest tags.
6. a video recommendations device, it is characterised in that including:
First list of videos generation module, for gathering the user behavior data of targeted customer, and according to described user behavior number
According to generating the first list of videos；
User characteristics vector generation module is for obtaining the user's characteristic information of described targeted customer and special according to described user
Reference breath generates user characteristics vector；
Forecast model determines module, for determining forecast model according to described first list of videos and described user characteristics vector；
Predictor calculation module, for calculating the predictive value of all videos to be selected according to described forecast model；
Video recommendations List Generating Module, for screening from all described videos to be selected according to the predictive value of described video to be selected
Go out video to be recommended, and generate video recommendations list according to described video to be recommended.
Device the most according to claim 6, it is characterised in that described first list of videos generation module includes:
User behavior data gathers submodule, all user behavior numbers of the described targeted customer in gathering the appointment time period
According to；
User behavior data screening submodule, for filtering out effective user behavior number from the user behavior data gathered
According to；
Sorting sub-module, for the time corresponding according to described effective user behavior data to described effective user behavior number
According to being ranked up, obtain described first list of videos.
Device the most according to claim 6, it is characterised in that described forecast model determines that module includes:
Supervision vector determines submodule, for determining supervision vector as shown in Equation 1 according to described user characteristics vector；
Wherein, uiRepresent i-th targeted customer, E (ui) represent that the supervision of described i-th targeted customer is vectorial, H (ui)kRepresent institute
State the kth user characteristics vector of i-th targeted customer, MkRepresent the kth user characteristics vector of described i-th targeted customer
Weight；
Forecast model determines submodule, is used for using shot and long term to remember recurrent neural network, according to described first list of videos and
Described supervision vector determines described forecast model.
9. according to the device described in claim 6 to 8 any one, it is characterised in that described predictor calculation module is specifically used
In:
Employing formula 2 calculates the predictive value of described video to be selected respectively；
Wherein, vjRepresent jth video to be selected, vtRepresent the t video in described first list of videos, uiRepresent i-th mesh
Mark user, p (vj|vt,ui) represent the predictive value of described jth video to be selected, s (vt,ui)jWith s (vt,ui)mAccording to formula 3 or formula 4
Determine；
Wherein, l represents the l neuron of the output layer of neutral net, v1Represent that the in described first list of videos the 1st regards
Frequently, t is more than 1, and n represents the n-th neuron of the hidden layer of described neutral net, lstm (v1,ui)n、lstm(vt-1,ui) and
lstm(vt,ui,lstm(vt-1,ui))nDetermine according to described forecast model, wlnRepresent described neutral net hidden layer n-th
Individual neuron is for the weight of l neuron of the output layer of described neutral net.
10. according to the device described in claim 6 to 8 any one, it is characterised in that described video recommendations List Generating Module
Including:
First screening submodule, for filtering out predictive value more than setting the described to be selected of threshold value from all described videos to be selected
Video, obtains the second list of videos；
Second screening submodule, for according to following at least one from described second list of videos, filter out described to be recommended regard
Frequently: channel information, described targeted customer belonging to the uploader information of described video to be selected, described video to be selected watch video
Data and the interest tags of described targeted customer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610561318.6A CN105975641A (en) | 2016-07-15 | 2016-07-15 | Video recommendation method ad device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610561318.6A CN105975641A (en) | 2016-07-15 | 2016-07-15 | Video recommendation method ad device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105975641A true CN105975641A (en) | 2016-09-28 |
Family
ID=56952696
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610561318.6A Pending CN105975641A (en) | 2016-07-15 | 2016-07-15 | Video recommendation method ad device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105975641A (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106507143A (en) * | 2016-10-21 | 2017-03-15 | 北京小米移动软件有限公司 | Video recommendation method and device |
| CN106897404A (en) * | 2017-02-14 | 2017-06-27 | 中国船舶重工集团公司第七0九研究所 | A kind of recommendation method and system based on many GRU layers of neutral nets |
| CN107507054A (en) * | 2017-07-24 | 2017-12-22 | 哈尔滨工程大学 | A kind of proposed algorithm based on Recognition with Recurrent Neural Network |
| CN107623866A (en) * | 2017-11-06 | 2018-01-23 | 深圳Tcl数字技术有限公司 | Video recommendation method, device and the storage medium of television set |
| WO2018076695A1 (en) * | 2016-10-31 | 2018-05-03 | 上海斐讯数据通信技术有限公司 | Smart recommendation system and smart recommendation method |
| CN108260008A (en) * | 2018-02-11 | 2018-07-06 | 北京未来媒体科技股份有限公司 | A kind of video recommendation method, device and electronic equipment |
| CN108304435A (en) * | 2017-09-08 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Information recommendation method, device, computer equipment and storage medium |
| CN108304115A (en) * | 2018-02-12 | 2018-07-20 | 广东欧珀移动通信有限公司 | Terminal entry selection method, device and storage medium and terminal |
| CN108460060A (en) * | 2017-02-22 | 2018-08-28 | 合网络技术（北京）有限公司 | Video recommendation method and device |
| CN108462900A (en) * | 2017-02-22 | 2018-08-28 | 合网络技术（北京）有限公司 | Video recommendation method and device |
| CN109104620A (en) * | 2018-07-26 | 2018-12-28 | 腾讯科技（深圳）有限公司 | A kind of short video recommendation method, device and readable medium |
| CN109729395A (en) * | 2018-12-14 | 2019-05-07 | 广州市百果园信息技术有限公司 | Video quality evaluation method, device, storage medium and computer equipment |
| CN109874053A (en) * | 2019-02-21 | 2019-06-11 | 南京航空航天大学 | The short video recommendation method with user's dynamic interest is understood based on video content |
| CN110069663A (en) * | 2019-04-29 | 2019-07-30 | 厦门美图之家科技有限公司 | Video recommendation method and device |
| WO2019233077A1 (en) * | 2018-06-08 | 2019-12-12 | 北京三快在线科技有限公司 | Ranking of business object |
| WO2020093781A1 (en) * | 2018-11-06 | 2020-05-14 | 北京达佳互联信息技术有限公司 | Multimedia resource estimated click through rate determination method and apparatus, and server |
| CN112822526A (en) * | 2020-12-30 | 2021-05-18 | 咪咕文化科技有限公司 | Video recommendation method, server and readable storage medium |
| CN112861001A (en) * | 2021-02-18 | 2021-05-28 | 咪咕数字传媒有限公司 | Method and device for generating recommendation value of digital content, electronic equipment and storage medium |
| CN114079826A (en) * | 2020-08-14 | 2022-02-22 | 北京达佳互联信息技术有限公司 | Video recommendation list generation method and device, server and storage medium |
| US11468536B2 (en) | 2018-05-18 | 2022-10-11 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for recommending a personalized pick-up location |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101694652A (en) * | 2009-09-30 | 2010-04-14 | 西安交通大学 | Network resource personalized recommended method based on ultrafast neural network |
| CN102929928A (en) * | 2012-09-21 | 2013-02-13 | 北京格致璞科技有限公司 | Multidimensional-similarity-based personalized news recommendation method |
| CN104679743A (en) * | 2013-11-26 | 2015-06-03 | 阿里巴巴集团控股有限公司 | Method and device for determining preference model of user |
| CN104731861A (en) * | 2015-02-05 | 2015-06-24 | 腾讯科技（深圳）有限公司 | Method and device for pushing multimedia data |
-
2016
- 2016-07-15 CN CN201610561318.6A patent/CN105975641A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101694652A (en) * | 2009-09-30 | 2010-04-14 | 西安交通大学 | Network resource personalized recommended method based on ultrafast neural network |
| CN102929928A (en) * | 2012-09-21 | 2013-02-13 | 北京格致璞科技有限公司 | Multidimensional-similarity-based personalized news recommendation method |
| CN104679743A (en) * | 2013-11-26 | 2015-06-03 | 阿里巴巴集团控股有限公司 | Method and device for determining preference model of user |
| CN104731861A (en) * | 2015-02-05 | 2015-06-24 | 腾讯科技（深圳）有限公司 | Method and device for pushing multimedia data |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106507143A (en) * | 2016-10-21 | 2017-03-15 | 北京小米移动软件有限公司 | Video recommendation method and device |
| WO2018076695A1 (en) * | 2016-10-31 | 2018-05-03 | 上海斐讯数据通信技术有限公司 | Smart recommendation system and smart recommendation method |
| CN106897404A (en) * | 2017-02-14 | 2017-06-27 | 中国船舶重工集团公司第七0九研究所 | A kind of recommendation method and system based on many GRU layers of neutral nets |
| CN106897404B (en) * | 2017-02-14 | 2021-04-09 | 中国船舶重工集团公司第七0九研究所 | Recommendation method and system based on multi-GRU layer neural network |
| CN108460060A (en) * | 2017-02-22 | 2018-08-28 | 合网络技术（北京）有限公司 | Video recommendation method and device |
| CN108462900B (en) * | 2017-02-22 | 2020-07-28 | 阿里巴巴（中国）有限公司 | Video recommendation method and device |
| CN108462900A (en) * | 2017-02-22 | 2018-08-28 | 合网络技术（北京）有限公司 | Video recommendation method and device |
| CN107507054A (en) * | 2017-07-24 | 2017-12-22 | 哈尔滨工程大学 | A kind of proposed algorithm based on Recognition with Recurrent Neural Network |
| WO2019047672A1 (en) * | 2017-09-08 | 2019-03-14 | 腾讯科技（深圳）有限公司 | Information recommendation method, computer device, and storage medium |
| US11514260B2 (en) | 2017-09-08 | 2022-11-29 | Tencent Technology (Shenzhen) Company Limited | Information recommendation method, computer device, and storage medium |
| CN108304435A (en) * | 2017-09-08 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Information recommendation method, device, computer equipment and storage medium |
| CN108304435B (en) * | 2017-09-08 | 2020-08-25 | 腾讯科技（深圳）有限公司 | Information recommendation method and device, computer equipment and storage medium |
| CN107623866B (en) * | 2017-11-06 | 2021-04-27 | 深圳Tcl数字技术有限公司 | Video recommendation method and device for television and storage medium |
| CN107623866A (en) * | 2017-11-06 | 2018-01-23 | 深圳Tcl数字技术有限公司 | Video recommendation method, device and the storage medium of television set |
| CN108260008A (en) * | 2018-02-11 | 2018-07-06 | 北京未来媒体科技股份有限公司 | A kind of video recommendation method, device and electronic equipment |
| CN108304115A (en) * | 2018-02-12 | 2018-07-20 | 广东欧珀移动通信有限公司 | Terminal entry selection method, device and storage medium and terminal |
| CN108304115B (en) * | 2018-02-12 | 2020-12-04 | Oppo广东移动通信有限公司 | Terminal item selection method and device, storage medium and terminal |
| US11468536B2 (en) | 2018-05-18 | 2022-10-11 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for recommending a personalized pick-up location |
| CN110647696A (en) * | 2018-06-08 | 2020-01-03 | 北京三快在线科技有限公司 | Business object sorting method and device |
| US20210366006A1 (en) * | 2018-06-08 | 2021-11-25 | Beijing Sankuai Online Technology Co., Ltd | Ranking of business object |
| WO2019233077A1 (en) * | 2018-06-08 | 2019-12-12 | 北京三快在线科技有限公司 | Ranking of business object |
| CN110647696B (en) * | 2018-06-08 | 2022-06-14 | 北京三快在线科技有限公司 | Business object sorting method and device |
| CN109104620A (en) * | 2018-07-26 | 2018-12-28 | 腾讯科技（深圳）有限公司 | A kind of short video recommendation method, device and readable medium |
| WO2020093781A1 (en) * | 2018-11-06 | 2020-05-14 | 北京达佳互联信息技术有限公司 | Multimedia resource estimated click through rate determination method and apparatus, and server |
| CN109729395A (en) * | 2018-12-14 | 2019-05-07 | 广州市百果园信息技术有限公司 | Video quality evaluation method, device, storage medium and computer equipment |
| CN109729395B (en) * | 2018-12-14 | 2022-02-08 | 广州市百果园信息技术有限公司 | Video quality evaluation method and device, storage medium and computer equipment |
| CN109874053B (en) * | 2019-02-21 | 2021-10-22 | 南京航空航天大学 | Short video recommendation method based on video content understanding and user dynamic interest |
| CN109874053A (en) * | 2019-02-21 | 2019-06-11 | 南京航空航天大学 | The short video recommendation method with user's dynamic interest is understood based on video content |
| CN110069663A (en) * | 2019-04-29 | 2019-07-30 | 厦门美图之家科技有限公司 | Video recommendation method and device |
| CN114079826A (en) * | 2020-08-14 | 2022-02-22 | 北京达佳互联信息技术有限公司 | Video recommendation list generation method and device, server and storage medium |
| CN114079826B (en) * | 2020-08-14 | 2023-11-28 | 北京达佳互联信息技术有限公司 | Video recommendation list generation method, device, server and storage medium |
| CN112822526A (en) * | 2020-12-30 | 2021-05-18 | 咪咕文化科技有限公司 | Video recommendation method, server and readable storage medium |
| CN112822526B (en) * | 2020-12-30 | 2023-07-07 | 咪咕文化科技有限公司 | Video recommendation method, server and readable storage medium |
| CN112861001A (en) * | 2021-02-18 | 2021-05-28 | 咪咕数字传媒有限公司 | Method and device for generating recommendation value of digital content, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105975641A (en) | Video recommendation method ad device | |
| CN107038213B (en) | Video recommendation method and device | |
| WO2021042826A1 (en) | Video playback completeness prediction method and apparatus | |
| CN108108821A (en) | Model training method and device | |
| CN106294830A (en) | The recommendation method and device of multimedia resource | |
| US9183497B2 (en) | Performance-efficient system for predicting user activities based on time-related features | |
| CN107306355B (en) | A kind of content recommendation method and server | |
| CN106168980A (en) | Multimedia resource recommends sort method and device | |
| CN108629665A (en) | A kind of individual commodity recommendation method and system | |
| CN110851706B (en) | Training method and device for user click model, electronic equipment and storage medium | |
| CN108319723A (en) | A kind of picture sharing method and device, terminal, storage medium | |
| CN112765373B (en) | Resource recommendation method and device, electronic equipment and storage medium | |
| CN109447713A (en) | A kind of recommended method and device of knowledge based map | |
| CN107404656A (en) | Live video recommends method, apparatus and server | |
| CN104053023B (en) | A kind of method and device of determining video similarity | |
| CN106604051A (en) | Live channel recommending method and device | |
| CN110442788A (en) | A kind of information recommendation method and device | |
| CN106326391A (en) | Method and device for recommending multimedia resources | |
| Hii et al. | Multigap: Multi-pooled inception network with text augmentation for aesthetic prediction of photographs | |
| CN108595493A (en) | Method for pushing and device, storage medium, the electronic device of media content | |
| CN106372101A (en) | Video recommendation method and apparatus | |
| CN108664654A (en) | A kind of main broadcaster's recommendation method and device based on user's similarity | |
| CN110781377B (en) | Article recommendation method and device | |
| CN114637911B (en) | Method for recommending next interest point of attention fusion perception network | |
| CN110335123A (en) | Method of Commodity Recommendation, system, computer-readable medium and device based on social electric business platform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160928 |
|
| RJ01 | Rejection of invention patent application after publication |