CN111400512A - Method and device for screening multimedia resources - Google Patents
Method and device for screening multimedia resources Download PDFInfo
- Publication number
- CN111400512A CN111400512A CN202010157401.3A CN202010157401A CN111400512A CN 111400512 A CN111400512 A CN 111400512A CN 202010157401 A CN202010157401 A CN 202010157401A CN 111400512 A CN111400512 A CN 111400512A
- Authority
- CN
- China
- Prior art keywords
- vector
- multimedia resource
- multimedia
- similarity
- account
- 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.)
- Granted
Links
- 238000012216 screening Methods 0.000 title claims abstract description 62
- 238000000034 method Methods 0.000 title claims abstract description 29
- 239000013598 vector Substances 0.000 claims abstract description 208
- 238000005259 measurement Methods 0.000 claims abstract description 24
- 230000000875 corresponding effect Effects 0.000 claims description 90
- 230000006870 function Effects 0.000 claims description 57
- 230000002596 correlated effect Effects 0.000 claims description 14
- 230000000694 effects Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 7
- 238000004590 computer program Methods 0.000 description 7
- 238000011156 evaluation Methods 0.000 description 6
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 230000006399 behavior Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 239000008267 milk Substances 0.000 description 3
- 210000004080 milk Anatomy 0.000 description 3
- 235000013336 milk Nutrition 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000001172 regenerating effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
- G06F16/43—Querying
- G06F16/435—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
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/23—Updating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the technical field of internet, in particular to a method and a device for screening multimedia resources. The method comprises the following steps: and respectively taking the pre-estimated feedback information corresponding to each multimedia resource as a new vector dimension, updating the corresponding multimedia resource vector, correspondingly updating the dimension of the account vector according to an updating result, respectively calculating the similarity between the updated account vector and each updated multimedia resource vector by adopting a preset measurement function, and screening out the multimedia resources corresponding to the multimedia resource vector meeting the preset condition based on the similarity to output as a screening result. Therefore, the multimedia resources meeting the requirements can be directly screened out through one-time retrieval in the recall stage, so that recall loss caused by the funnel effect is effectively avoided, the screening accuracy is improved, the retrieval time is shortened, the screening efficiency of the multimedia resources is improved, and the computing resources are saved.
Description
Technical Field
The present application relates to the field of internet technologies, and in particular, to a method and an apparatus for screening multimedia resources.
Background
The development of deep learning technology is promoting the deep revolution of various industries. In the multimedia retrieval system, the deep learning technique is mainly applied to the behavior estimation of click rate, conversion rate, etc. In a large-scale multimedia search system, the number of multimedia resources that can be selected to be released under each account is huge, and it is impossible to select the most suitable multimedia resources by one-time calculation, so the most suitable multimedia resources are generally selected by adopting a multi-level funnel mode, for example, through three main stages of recall, rough selection and fine selection, the multimedia resources adapted to be played in a certain account are selected. By adopting the method, the multimedia resources suitable for presentation can be screened out aiming at different accounts, and finally, the benefit maximization is realized among the accounts, the multimedia resource provider and the playing platform, and the mutual profit and win-win are realized.
In the prior art, in the recall stage, the number of candidate multimedia resources is huge, so that a great amount of computing resources are needed for estimating user behaviors by using a deep learning model, and how to obtain higher computing precision in a limited time is an important index of a multimedia resource retrieval system.
The current solution is mainly divided into two phases:
stage one: and (3) quantifying the account and the multimedia resource into a low-dimensional continuous vector in the same vector space by adopting a double-tower model through a neural network. The quantization process can be completed on line, and the multimedia resource vector which is relatively close to the user vector similarity can be searched on line only through the account vector.
And a second stage: in the obtained multimedia resources with higher similarity, based on the estimated feedback information of each multimedia resource, further calculating recommendation evaluation information (which may be recorded as cpm) of each multimedia resource, so as to further screen out a batch of multimedia resources with higher recommendation evaluation information, and recommending the multimedia resources as candidate multimedia resources screened out in a recall stage, wherein the estimated feedback information refers to a feedback operation behavior which can be obtained by estimation after the multimedia resources are played, and optionally, the estimated feedback information may be bid, may be a conversion rate after viewing (cvr), may also be an app activation rate, and the like.
The following description will take multimedia resources as advertisements and the estimated feedback information as the evaluation of the advertiser as an example.
Supposing that, aiming at an account X, the system firstly screens out a batch of advertisement sets with higher ctr values, which are marked as S1, ctr represents the similarity between a multimedia resource vector and an account vector, and then further screens out a batch of advertisement sets with higher cpm from S1 by adopting a formula cpm (ctr) bid, which is marked as S, wherein S is the advertisement set with higher similarity to the account X, and meanwhile, the advertiser is estimated to give out higher advertisement evaluation.
It can be seen that the operation flow of generating S1 and regenerating S in the recall stage increases the complexity of execution, thereby reducing the screening efficiency, and further, due to the deviation of the prediction target, S1 and S are not generally the same set, so there is a loss of funnel efficiency, thereby reducing the screening accuracy.
In view of the above, a new technical solution for screening multimedia resources needs to be designed to overcome the above-mentioned drawbacks.
Disclosure of Invention
The embodiment of the invention provides a method and a device for screening multimedia resources, which are used for improving the screening efficiency of the multimedia resources and the screening accuracy in a recall stage.
The embodiment of the invention provides the following specific technical scheme:
in a first aspect, a method for screening multimedia resources includes:
acquiring an account vector of a target account and a multimedia resource vector of each multimedia resource to be screened;
respectively taking the pre-estimated feedback information corresponding to each multimedia resource as a new vector dimension, updating the corresponding multimedia resource vector, and correspondingly updating the dimension of the account vector according to the updating result;
respectively calculating the similarity between the updated account vector and each updated multimedia resource vector by adopting a preset measurement function, wherein the similarity is positively correlated with the estimated feedback information contained in one updated multimedia resource vector;
and screening out the multimedia resources corresponding to the multimedia resource vectors with the similarity meeting the preset conditions based on the obtained similarity, and outputting the multimedia resources as screening results.
Optionally, the updating of the corresponding multimedia resource vector by using the pre-estimated feedback information corresponding to one multimedia resource as a new vector dimension includes:
taking the pre-estimated feedback information corresponding to the multimedia resource as a true number, taking a constant e as a base number, and generating a corresponding new vector dimension by adopting a logarithmic function;
and adding the new vector dimension to the multimedia resource vector.
Optionally, the correspondingly updating the dimension of the account vector according to the update result includes:
adding a new dimension with a value of 1 to the account vector.
Optionally, a preset measurement function is adopted, and a similarity between the updated account vector and an updated multimedia resource vector is calculated respectively, wherein the similarity is positively correlated with the estimated feedback information contained in the updated multimedia resource vector, and includes:
determining a preset measurement function, wherein the measurement function at least comprises an exponential function taking a constant e as a base number;
and calculating to obtain corresponding similarity by using the exponential function and taking the product of the updated account vector and the updated multimedia resource vector as an index.
Optionally, based on the obtained similarities, screening out the multimedia resources corresponding to the multimedia resource vectors whose similarities meet the preset condition, including:
screening out multimedia resources corresponding to each multimedia resource vector with the similarity reaching a set threshold; alternatively, the first and second electrodes may be,
and screening out the multimedia resources corresponding to the N multimedia resource vectors with the highest similarity, wherein N is a preset natural number.
In a second aspect, an apparatus for screening multimedia resources includes:
the acquisition unit is used for acquiring an account vector of the target account and multimedia resource vectors of various multimedia resources to be screened;
the updating unit is used for respectively taking the pre-estimated feedback information corresponding to each multimedia resource as a new vector dimension, updating the corresponding multimedia resource vector and correspondingly updating the dimension of the account vector according to the updating result;
the computing unit is used for respectively computing the similarity between the updated account vector and each updated multimedia resource vector by adopting a preset measurement function, wherein the similarity is positively correlated with the estimated feedback information contained in one updated multimedia resource vector;
and the screening unit is used for screening out the multimedia resources corresponding to the multimedia resource vectors with the similarity meeting the preset conditions based on the obtained similarities, and outputting the multimedia resources as screening results.
Optionally, the estimated feedback information corresponding to one multimedia resource is used as a new vector dimension, and when the corresponding multimedia resource vector is updated, the updating unit is configured to:
taking the pre-estimated feedback information corresponding to the multimedia resource as a true number, taking a constant e as a base number, and generating a corresponding new vector dimension by adopting a logarithmic function;
and adding the new vector dimension to the multimedia resource vector.
Optionally, when the dimension of the account vector is updated according to the update result, the update unit is configured to:
adding a new dimension with a value of 1 to the account vector.
Optionally, a preset measurement function is adopted, and a similarity between the updated account vector and an updated multimedia resource vector is calculated respectively, wherein when the similarity is positively correlated with the estimated feedback information contained in an updated multimedia resource vector, the calculation unit is configured to:
determining a preset measurement function, wherein the measurement function at least comprises an exponential function with a constant e as a constant;
and calculating to obtain corresponding similarity by using the exponential function and taking the product of the updated account vector and the updated multimedia resource vector as an index.
Optionally, when the multimedia resource corresponding to the multimedia resource vector whose similarity meets the preset condition is screened out based on the obtained similarities, the screening unit is configured to:
screening out multimedia resources corresponding to each multimedia resource vector with the similarity reaching a set threshold; alternatively, the first and second electrodes may be,
and screening out the multimedia resources corresponding to the N multimedia resource vectors with the highest similarity, wherein N is a preset natural number.
In a third aspect, a network device includes:
a memory for storing executable instructions;
a processor for reading and executing executable instructions stored in the memory to implement the method of any one of the above first aspects.
In a fourth aspect, a storage medium, wherein instructions, when executed by a processor, enable the processor to perform the method of any of the first aspect.
In the embodiment of the application, the pre-estimated feedback information corresponding to each multimedia resource is used as a new vector dimension to update the corresponding multimedia resource vector, the dimension of the account vector is correspondingly updated according to the updating result, and the similarity between the updated account vector and each updated multimedia resource vector is respectively calculated by adopting a preset measurement function, wherein the similarity is positively correlated with the pre-estimated feedback information contained in one updated multimedia resource vector, and based on the similarity, the multimedia resource corresponding to the multimedia resource vector meeting the preset condition can be screened out and output as the screening result. Therefore, the multimedia resources meeting the requirements can be directly screened out through one-time retrieval in the recall stage, so that recall loss caused by the funnel effect is effectively avoided, the screening accuracy is improved, the retrieval time is shortened, the screening efficiency of the multimedia resources is improved, and the computing resources are saved.
Drawings
FIG. 1 is a schematic flow chart illustrating the resource multimedia resource in the recall stage according to an embodiment of the present application;
FIG. 2 is a diagram illustrating a function image of a constant c in an embodiment of the present application;
FIG. 3 is a schematic diagram of a logical architecture of a network device according to an embodiment of the present application;
fig. 4 is a schematic entity architecture diagram of a network device in the embodiment of the present application.
Detailed Description
In order to improve the screening efficiency of the multimedia resources and improve the screening accuracy in the recall stage, in the embodiment of the application, the pre-estimated feedback information of the multimedia resources is respectively converted into one dimension of an account vector and a multimedia resource vector, so that the multimedia resources needing to be recommended in the recall stage can be directly screened out only through once similarity calculation.
Preferred embodiments of the present application will be described in further detail below with reference to the accompanying drawings.
In practical application, the multimedia resource retrieval system can screen out appropriate multimedia resources for recommendation respectively aiming at each account, wherein the screening process comprises three stages of recall, rough selection and fine selection. In the following embodiments, only one account is taken as an example, and the screening process in the recall stage is described in detail.
Referring to fig. 1, in the embodiment of the present application, a specific process of screening multimedia resources in a recall phase is as follows:
step 100: and acquiring an account vector of the target account and a multimedia resource vector of each multimedia resource to be screened.
Specifically, an existing double-tower model may be adopted, and a corresponding initial account vector may be generated on the basis of a target account on line and recorded as U _ ctr, and a corresponding initial multimedia resource vector may be generated on the basis of each multimedia resource on line and recorded as V _ ctr.
Wherein, each dimension in U _ ctr represents an attribute of the target account, such as: { user age, point of interest, payroll level … … };
each dimension in V _ ctr represents an attribute of a multimedia asset vector, such as { industry type, target user, Play scope, … … }.
In practical applications, the physical meanings of the dimensions in the U _ ctr and V _ ctr vectors are implicit and are both learned by using the model, that is, the model inputs the high latitude vector with explicit meaning, the model outputs the low latitude vector with implicit meaning, and U _ ctr and V _ ctr refer to the latter, so that the above description is only a schematic description, and is not an actual implementation manner, and specific examples are given in the following embodiments, and thus, the emphasis is placed.
Step 110: and respectively taking the pre-estimated feedback information corresponding to each multimedia resource as a new vector dimension, updating the corresponding multimedia resource vector, and correspondingly updating the dimension of the account vector according to the updating result.
Specifically, the estimated feedback information corresponding to one multimedia resource represents the feedback operation behavior that can be obtained by estimation after the multimedia resource is played, so that the step 110 is executed to determine the bid corresponding to each multimedia resource.
For example, if a multimedia resource is a milk advertisement, the bid corresponding to the milk advertisement may be an average value of advertisement evaluations given by each advertiser after the milk advertisement is played.
For another example, assuming that one multimedia resource is a lipstick advertisement, the bid corresponding to the lipstick advertisement may be an online sales amount of lipstick within a set time period (e.g., one day) after the lipstick advertisement is played.
In short, the bid corresponding to one multimedia resource represents the social benefit generated after the one multimedia resource is played.
Further, taking a multimedia resource as an example, when a multimedia resource vector is updated, a bid corresponding to the multimedia resource may be added to a corresponding multimedia resource vector as a new vector dimension, and optionally, the updated multimedia resource vector may be represented as:
V_cpm＝concat(V_ctr，loge (bid)log(bid))
it can be seen from the above description that in the embodiment of the present application, bid corresponding to one multimedia resource is used as a true number, a constant e is used as a base number, a new dimension is generated in a logarithmic function form, and the new dimension is added to a multimedia resource vector corresponding to the one multimedia resource, so that an update operation is completed.
The logarithmic form is used to correspond to the metric function f (x) used later, as will be explained in the following embodiments.
Correspondingly, since one dimension is added to the multimedia resource vector, in order to facilitate subsequent similarity calculation, a corresponding dimension needs to be added to the account vector of the target account to update the account vector, and optionally, the updated account vector may be:
U_cpm＝concat(U_ctr，1)
it can be seen from the above that, in the embodiment of the present application, the value "1" is directly adopted as a new dimension, and is added to the account vector, so that the update operation is completed.
Step 120: and respectively calculating the similarity between the updated account vector and each updated multimedia resource vector by adopting a preset measurement function, wherein the similarity is positively correlated with the estimated feedback information contained in one updated multimedia resource vector.
Optionally, in the embodiment of the present application, a new measurement function f (x) is designed, and a specific expression of f (x) is as follows:
f(U,V)＝min(eU.V,(1-c2)·sigmoid(U,V)+c2),c∈[0,1]
from the above formula, it can be seen thatIn the embodiment of the present application, an exponential function e is adoptedU.VThe similarity f (x) is calculated, and V _ cpm ═ concat (V _ ctr, log (bid)), since the logarithmic function and the exponential function are inverse functions to each other, V _ cpm is substituted into the above eU·VLater, it can be seen that f (U, V) is positively correlated with bid, so the value of f (U, V) can be influenced by adjusting bid.
In the above formula, c is a preset constant between 0 and 1, which may be generally 0.9, and the function image is shown in fig. 2.
In general, f (U, V) (i.e., similarity) has a value range of (0, 1), and the output can be directly used as a probability value, i.e., the probability that the target account clicks and plays after recommending a multimedia resource.
And in [ - ∞, log (c)]Within the interval (e) is an exponential function eU·VThe property of the exponential function is that the addition of the multiplication conversion input can be output, and the general probability values are all located at 0, c when the values are larger]Interval, so that the normal value of U.V is [ - ∞, log (c)]In the meantime. In other words, if the exponential function eU·VThe calculated value is between (0, 1), then the exponential function e is usedU·VOtherwise, use (1-c)2)·sigmoid(U,V)+c2) The calculation result of (2) is used to control the value of the probability value between (0, 1), specifically, (1-c) may not be used2)·sigmoid(U,V)+c2) The calculation is performed while other formulas are used, and this is only an example.
Furthermore, in the above embodiment, the expression of V _ cpm, the expression of U _ cpm, and the expression of f (U, V) are all examples, and in practical application, flexible configuration may be performed according to an application scenario, only the value of bid needs to be added to the multimedia resource vector as a new dimension, and meanwhile, the expression of f (U, V) also needs to be associated with the value of bid, so that retrieval is performed by using V _ cpm and U _ cpm, and a multimedia resource vector meeting requirements can be directly screened out from the multimedia resource vector at one time, thereby obtaining a corresponding multimedia resource set S.
Step 130: and screening out the multimedia resources corresponding to the multimedia resource vectors with the similarity meeting the preset conditions based on the obtained similarity, and outputting the multimedia resources as screening results.
Specifically, when step 130 is executed, the following two ways may be adopted, but not limited to:
the first mode is as follows: and screening out the multimedia resources corresponding to the multimedia resource vectors with the similarity reaching the set threshold.
For example, if the set threshold is 80%, multimedia resources corresponding to the updated multimedia resource vector are screened from the recalled multimedia resources, where the similarity between the multimedia resources and the updated account vector of the target account reaches 80%, and a corresponding multimedia resource set S is generated.
Therefore, all multimedia resources meeting the requirements can be screened out in the recall stage, and high-quality materials are provided for the subsequent rough selection stage and the fine selection stage so as to be further processed.
The second mode is as follows: and screening out the multimedia resources corresponding to the N multimedia resource vectors with the highest similarity, wherein N is a preset natural number.
For example, if N is 5, then 5 multimedia resources with the highest similarity value with the updated account vector of the target account are screened out from the recalled multimedia resources, and the multimedia resource corresponding to the updated multimedia resource vector is generated as the corresponding multimedia resource set S.
Therefore, the multimedia resources can be screened out as much as possible in the recall stage, and all materials which can meet the requirements are provided for the subsequent rough selection stage and the fine selection stage so as to be further processed.
Therefore, the estimated feedback information is integrated into the process of vector retrieval (namely similarity calculation) in a lossless manner by expanding the dimension of the account vector and expanding the dimension of the multimedia resource vector, and the loss caused by calculating ctr and then cpm is avoided.
The above embodiments are further described in detail with a specific application scenario.
Assume that the account vector corresponding to the target account is: u _ ctr ═ { 0.003542910.808438660.53149480.19999890.265827480.152742670.173107920.68672190.06591270.82318366 }, and accordingly, the updated account vector is: u _ cpm ═ 0.003542910.808438660.53149480.19999890.265827480.152742670.173107920.68672190.06591270.82318366, 1.
And, assuming that 5 multimedia resources are obtained in the recall phase, referred to as resource 1, resource 2, resource 3, resource 4, and resource 5, their respective multimedia resource vectors are:
V_ctr1＝{-2.05182582 -1.6897268 -2.33889794 -2.00365975 -1.81088718 -1.65403273 -1.83973446 -2.38171944 -1.9649727 -1.64757683}；
V_ctr2＝{-2.71072205 -2.36611976 -2.56170478 -2.48215594 -3.02894542-2.5466344 -2.5221289 -3.20289881 -3.09769717 -2.4181483}；
V_ctr3＝{-0.36522338 0.23268305 0.46526592 0.37448625 0.10760550.49859606 -0.08238123 0.12145822 -0.19842657 -0.1045822}；
V_ctr4＝{-1.99248004 -1.89444956 -1.56142056 -1.96741536 -2.02441806-1.5403093 -1.39481241 -1.92756832 -2.03972335 -1.8894462}；
V_ctr5＝{-0.20616983 -0.13013617 -0.83665253 -0.56258492 -0.52868513-0.40561355 -0.12736182 -0.63233977 0.06294681 -0.57332818}；
suppose that the estimated feedback information of the 5 multimedia resources is: bid 1-0.5612, bid 2-0.0998, bid 3-0.02369, bid 4-0.2165 and bid 5-0.0075.
Accordingly, each updated multimedia resource vector is:
V_cpm1＝{-2.05182582 -1.6897268 -2.33889794 -2.00365975 -1.81088718 -1.65403273 -1.83973446 -2.38171944 -1.9649727 -1.64757683 -0.5776}；
V_cpm2＝{-2.71072205 -2.36611976 -2.56170478 -2.48215594 -3.02894542-2.5466344 -2.5221289 -3.20289881 -3.09769717 -2.4181483 -2.304}；
V_cpm3＝{-0.36522338 0.23268305 0.46526592 0.37448625 0.10760550.49859606 -0.08238123 0.12145822 -0.19842657 -0.1045822 -3.742}；
V_cpm4＝{-1.99248004 -1.89444956 -1.56142056 -1.96741536 -2.02441806-1.5403093 -1.39481241 -1.92756832 -2.03972335 -1.8894462 -1.529}；
V_cpm5＝{-0.20616983 -0.13013617 -0.83665253 -0.56258492 -0.52868513-0.40561355 -0.12736182 -0.63233977 0.06294681 -0.57332818 -4.885}。
further, the similarity between each updated multimedia resource vector and the updated account vector is calculated using the above formula f (U, V), assuming the following results,
f(U_cpm1,V_cpm)＝0.0004；
f(U_cpm2,V_cpm)＝0.0；
f(U_cpm3,V_cpm)＝0.0425；
f(U_cpm4,V_cpm)＝0.0002；
f(U_cpm5,V_cpm)＝0.0013；
thus, after screening, f (U)_cpm2,V_cpm) And f (U)_cpm5,V_cpm) The two resources with the largest evaluation value are selected, and therefore, the multimedia resources finally selected in the recall stage are: resources 3 and 5, i.e. a set S of multimedia resources may be generated based on resources 3 and 5.
In the above process, 5 multimedia resources are used for example, and in practical application, the number of multimedia resources is much more than that of the above example, but all multimedia resources can be processed in the same manner, which is not described herein again.
Based on the same inventive concept, referring to fig. 3, an embodiment of the present application provides a network device, which at least includes an obtaining unit 30, an updating unit 31, a calculating unit 32, and a filtering unit 33, wherein,
an obtaining unit 30, configured to obtain an account vector of a target account and a multimedia resource vector of each multimedia resource to be screened;
the updating unit 31 is configured to update the corresponding multimedia resource vector by using the estimated feedback information corresponding to each multimedia resource as a new vector dimension, and correspondingly update the dimension of the account vector according to an update result;
the calculating unit 32 is configured to calculate, by using a preset metric function, a similarity between each updated account vector and each updated multimedia resource vector, where the similarity is positively correlated with the estimated feedback information included in one updated multimedia resource vector;
and the screening unit 33 is configured to screen out, based on the obtained similarity, a multimedia resource corresponding to the multimedia resource vector whose similarity meets a preset condition, and output the multimedia resource as a screening result.
Optionally, the estimated feedback information corresponding to one multimedia resource is used as a new vector dimension, and when the corresponding multimedia resource vector is updated, the updating unit 31 is configured to:
taking the pre-estimated feedback information corresponding to the multimedia resource as a true number, taking a constant e as a base number, and generating a corresponding new vector dimension by adopting a logarithmic function;
and adding the new vector dimension to the multimedia resource vector.
Optionally, when the dimension of the account vector is updated according to the update result, the updating unit 31 is configured to:
adding a new dimension with a value of 1 to the account vector.
Optionally, a preset measurement function is adopted to calculate a similarity between the updated account vector and an updated multimedia resource vector, where, when the similarity is positively correlated with the estimated feedback information contained in an updated multimedia resource vector, the calculating unit 32 is configured to:
determining a preset measurement function, wherein the measurement function at least comprises an exponential function with a constant e as a constant;
and calculating to obtain corresponding similarity by using the exponential function and taking the product of the updated account vector and the updated multimedia resource vector as an index.
Optionally, when the multimedia resource corresponding to the multimedia resource vector with the similarity meeting the preset condition is screened out based on the obtained similarities, the screening unit 33 is configured to:
screening out multimedia resources corresponding to each multimedia resource vector with the similarity reaching a set threshold; alternatively, the first and second electrodes may be,
and screening out the multimedia resources corresponding to the N multimedia resource vectors with the highest similarity, wherein N is a preset natural number.
Based on the same inventive concept, referring to fig. 4, an embodiment of the present application provides a network device, which at least includes:
a memory 402 for storing executable instructions;
a processor 401, configured to read and execute the executable instructions stored in the memory, so as to implement the following processes:
acquiring an account vector of a target account and a multimedia resource vector of each multimedia resource to be screened;
respectively taking the pre-estimated feedback information corresponding to each multimedia resource as a new vector dimension, updating the corresponding multimedia resource vector, and correspondingly updating the dimension of the account vector according to the updating result;
respectively calculating the similarity between the updated account vector and each updated multimedia resource vector by adopting a preset measurement function, wherein the similarity is positively correlated with the estimated feedback information contained in one updated multimedia resource vector;
and screening out the multimedia resources corresponding to the multimedia resource vectors with the similarity meeting the preset conditions based on the obtained similarity, and outputting the multimedia resources as screening results.
Optionally, the estimated feedback information corresponding to one multimedia resource is used as a new vector dimension, and when the corresponding multimedia resource vector is updated, the processor 401 is configured to:
taking the pre-estimated feedback information corresponding to the multimedia resource as a true number, taking a constant e as a base number, and generating a corresponding new vector dimension by adopting a logarithmic function;
and adding the new vector dimension to the multimedia resource vector.
Optionally, when the dimension of the account vector is updated according to the update result, the processor 401 is configured to:
adding a new dimension with a value of 1 to the account vector.
Optionally, a preset measurement function is adopted to calculate a similarity between the updated account vector and an updated multimedia resource vector, where, when the similarity is positively correlated with the estimated feedback information contained in an updated multimedia resource vector, the processor 401 is configured to:
determining a preset measurement function, wherein the measurement function at least comprises an exponential function with a constant e as a constant;
and calculating to obtain corresponding similarity by using the exponential function and taking the product of the updated account vector and the updated multimedia resource vector as an index.
Optionally, when the multimedia resource corresponding to the multimedia resource vector whose similarity meets the preset condition is screened out based on the obtained similarities, the processor 401 is configured to:
screening out multimedia resources corresponding to each multimedia resource vector with the similarity reaching a set threshold; alternatively, the first and second electrodes may be,
and screening out the multimedia resources corresponding to the N multimedia resource vectors with the highest similarity, wherein N is a preset natural number.
Based on the same inventive concept, embodiments of the present application provide a storage medium, and instructions in the storage medium, when executed by a processor, enable the processor to perform the method according to any one of the above embodiments.
To sum up, in the embodiment of the present application, the estimation feedback information that corresponds each multimedia resource is regarded as new vector dimension respectively, updates corresponding multimedia resource vector, and is right according to the result of updating the dimension of account vector carries out corresponding update to adopt predetermined measurement function, calculate the similarity between the account vector after the update and each multimedia resource vector after the update respectively, wherein, the similarity is positive correlation with the estimation feedback information who contains in the multimedia resource vector after an update, and, based on the similarity, alright select the multimedia resource that corresponds the multimedia resource vector who accords with the predetermined condition, export as the screening result. Therefore, the multimedia resources meeting the requirements can be directly screened out through one-time retrieval in the recall stage, so that recall loss caused by the funnel effect is effectively avoided, the screening accuracy is improved, the retrieval time is shortened, the screening efficiency of the multimedia resources is improved, and the computing resources are saved.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made in the embodiments of the present invention without departing from the spirit or scope of the embodiments of the invention. Thus, if such modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to encompass such modifications and variations.
Claims (10)
1. A method for screening multimedia resources, comprising:
acquiring an account vector of a target account and a multimedia resource vector of each multimedia resource to be screened;
respectively taking the pre-estimated feedback information corresponding to each multimedia resource as a new vector dimension, updating the corresponding multimedia resource vector, and correspondingly updating the dimension of the account vector according to the updating result;
respectively calculating the similarity between the updated account vector and each updated multimedia resource vector by adopting a preset measurement function, wherein the similarity is positively correlated with the estimated feedback information contained in one updated multimedia resource vector;
and screening out the multimedia resources corresponding to the multimedia resource vectors with the similarity meeting the preset conditions based on the obtained similarity, and outputting the multimedia resources as screening results.
2. The method of claim 1, wherein updating the corresponding multimedia resource vector using the estimated feedback information corresponding to the multimedia resource as a new vector dimension comprises:
taking the pre-estimated feedback information corresponding to the multimedia resource as a true number, taking a constant e as a base number, and generating a corresponding new vector dimension by adopting a logarithmic function;
and adding the new vector dimension to the multimedia resource vector.
3. The method of claim 2, wherein updating the dimensions of the account vector accordingly based on the update results comprises:
adding a new dimension with a value of 1 to the account vector.
4. The method according to claim 2 or 3, wherein the step of calculating the similarity between the updated account vector and an updated multimedia resource vector respectively by using a predetermined metric function, wherein the similarity is positively correlated with the estimated feedback information contained in the updated multimedia resource vector comprises:
determining a preset measurement function, wherein the measurement function at least comprises an exponential function taking a constant e as a base number;
and calculating to obtain corresponding similarity by using the exponential function and taking the product of the updated account vector and the updated multimedia resource vector as an index.
5. The method as claimed in claim 1, 2 or 3, wherein screening out multimedia resources corresponding to multimedia resource vectors with similarity meeting preset conditions based on the obtained respective similarities comprises:
screening out multimedia resources corresponding to each multimedia resource vector with the similarity reaching a set threshold; alternatively, the first and second electrodes may be,
and screening out the multimedia resources corresponding to the N multimedia resource vectors with the highest similarity, wherein N is a preset natural number.
6. An apparatus for screening multimedia resources, comprising:
the acquisition unit is used for acquiring an account vector of the target account and multimedia resource vectors of various multimedia resources to be screened;
the updating unit is used for respectively taking the pre-estimated feedback information corresponding to each multimedia resource as a new vector dimension, updating the corresponding multimedia resource vector and correspondingly updating the dimension of the account vector according to the updating result;
the computing unit is used for respectively computing the similarity between the updated account vector and each updated multimedia resource vector by adopting a preset measurement function, wherein the similarity is positively correlated with the estimated feedback information contained in one updated multimedia resource vector;
and the screening unit is used for screening out the multimedia resources corresponding to the multimedia resource vectors with the similarity meeting the preset conditions based on the obtained similarities, and outputting the multimedia resources as screening results.
7. The apparatus of claim 6, wherein the update unit is configured to, when updating the corresponding multimedia resource vector by using the estimated feedback information corresponding to one multimedia resource as a new vector dimension:
taking the pre-estimated feedback information corresponding to the multimedia resource as a true number, taking a constant e as a base number, and generating a corresponding new vector dimension by adopting a logarithmic function;
and adding the new vector dimension to the multimedia resource vector.
8. The apparatus of claim 7, wherein when the dimension of the account vector is updated accordingly according to the update result, the update unit is configured to:
adding a new dimension with a value of 1 to the account vector.
9. A network device, comprising:
a memory for storing executable instructions;
a processor for reading and executing executable instructions stored in the memory to implement the method of any one of claims 1 to 5.
10. A storage medium, wherein instructions in the storage medium, when executed by a processor, enable the processor to perform the method of any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010157401.3A CN111400512B (en) | 2020-03-09 | 2020-03-09 | Method and device for screening multimedia resources |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010157401.3A CN111400512B (en) | 2020-03-09 | 2020-03-09 | Method and device for screening multimedia resources |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111400512A true CN111400512A (en) | 2020-07-10 |
| CN111400512B CN111400512B (en) | 2023-12-01 |
Family
ID=71436147
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010157401.3A Active CN111400512B (en) | 2020-03-09 | 2020-03-09 | Method and device for screening multimedia resources |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111400512B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113420056A (en) * | 2021-05-14 | 2021-09-21 | 北京达佳互联信息技术有限公司 | Behavior data processing method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102473183A (en) * | 2009-07-14 | 2012-05-23 | 索尼公司 | Content recommendation system, content recommendation method, content recommendation device, and information recording medium |
| CN103207914A (en) * | 2013-04-16 | 2013-07-17 | 武汉理工大学 | Preference vector generation method and preference vector generation system based on user feedback evaluation |
| US20180330192A1 (en) * | 2017-05-15 | 2018-11-15 | International Business Machines Corporation | Load-Balancing Training of Recommender System for Heterogeneous Systems |
| CN110162703A (en) * | 2019-05-13 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Content recommendation method, training method, device, equipment and storage medium |
-
2020
- 2020-03-09 CN CN202010157401.3A patent/CN111400512B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102473183A (en) * | 2009-07-14 | 2012-05-23 | 索尼公司 | Content recommendation system, content recommendation method, content recommendation device, and information recording medium |
| CN103207914A (en) * | 2013-04-16 | 2013-07-17 | 武汉理工大学 | Preference vector generation method and preference vector generation system based on user feedback evaluation |
| US20180330192A1 (en) * | 2017-05-15 | 2018-11-15 | International Business Machines Corporation | Load-Balancing Training of Recommender System for Heterogeneous Systems |
| CN110162703A (en) * | 2019-05-13 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Content recommendation method, training method, device, equipment and storage medium |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113420056A (en) * | 2021-05-14 | 2021-09-21 | 北京达佳互联信息技术有限公司 | Behavior data processing method and device, electronic equipment and storage medium |
| CN113420056B (en) * | 2021-05-14 | 2023-12-26 | 北京达佳互联信息技术有限公司 | Behavior data processing method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111400512B (en) | 2023-12-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108540826B (en) | Bullet screen pushing method and device, electronic equipment and storage medium | |
| KR102066773B1 (en) | Method, apparatus and system for content recommendation | |
| US9864951B1 (en) | Randomized latent feature learning | |
| KR102203253B1 (en) | Rating augmentation and item recommendation method and system based on generative adversarial networks | |
| CN107463580B (en) | Click rate estimation model training method and device and click rate estimation method and device | |
| CN106599047B (en) | Information pushing method and device | |
| CN108205775A (en) | The recommendation method, apparatus and client of a kind of business object | |
| CN111209477A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN111783873A (en) | Incremental naive Bayes model-based user portrait method and device | |
| CN113239182A (en) | Article recommendation method and device, computer equipment and storage medium | |
| CN113435499A (en) | Label classification method and device, electronic equipment and storage medium | |
| CN113407854A (en) | Application recommendation method, device and equipment and computer readable storage medium | |
| CN112182281B (en) | Audio recommendation method, device and storage medium | |
| CN115471271A (en) | Method and device for attributing advertisements, computer equipment and readable storage medium | |
| JP2007323315A (en) | Cooperative filtering method, cooperative filtering device, cooperative filtering program and recording medium with the same program recorded thereon | |
| EP4171045A1 (en) | Production method and device for multimedia works, and computer-readable storage medium | |
| CN110807693A (en) | Album recommendation method, device, equipment and storage medium | |
| CN111160638A (en) | Conversion estimation method and device | |
| CN111400512B (en) | Method and device for screening multimedia resources | |
| CN116757783A (en) | Product recommendation method and device | |
| CN113836388A (en) | Information recommendation method and device, server and storage medium | |
| CN113836390A (en) | Resource recommendation method and device, computer equipment and storage medium | |
| CN110569447A (en) | network resource recommendation method and device and storage medium | |
| CN112269942B (en) | Method, device and system for recommending object and electronic equipment | |
| CN115017413A (en) | Recommendation method and device, computing equipment and computer storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |