CN110889063A - Video pre-caching method based on Hox process and matrix decomposition - Google Patents

Abstract

The invention provides a video pre-caching method based on a Hox process and matrix decomposition, which comprises the following steps: s1, expressing the proportion of cached requests in all the requests in a period of time by using a hit rate, and mathematically defining the hit rate again by using a mode suitable for a point process according to a hit rate model; s2, predicting the intensity of all videos and sequencing the videos according to the self history of the equipment by using a self-excited Hox process caching model, and caching the videos with a certain size; s3, adding influences of neighbor historical clicks on equipment by a cache model in the process of mutually exciting the Hox, predicting the strength of all videos, and improving the hit rate; s4, carrying out decomposition and dimension reduction on the parameters of the S3 by using a matrix decomposition and dimension reduction model; and S5, optimizing the mutual excitation Hox process cache model by using a process optimization algorithm, and predicting the future behavior intensity. The method and the device can effectively predict the watching intensity of each device to different videos in the future, and can remarkably reduce the flow in the future network and improve the user experience by pre-caching some videos with the maximum intensity.

Description

Video pre-caching method based on Hox process and matrix decomposition

Technical Field

The invention relates to the field of network video caching, in particular to a video pre-caching method based on a Hox process and matrix decomposition.

Background

With the development of online video playing market, online video playing is more and more popular, and the number of people watching online videos is more and more. Cisco predicts that in the future, most of the Internet traffic will be video traffic, and at the mobile end, the proportion of the Internet traffic generated by the online video service will increase from 59% in 2017 to 79% in 2022. At the same time, the number of online videos is also growing at an alarming rate, statistically, 300 hours of video content are uploaded to YouTube each day, including UGC, news, drama, movie, etc. Accordingly, there are more and more transmission or terminal devices in the internet, so to relieve the heavy traffic load of online video, the video content can be cached in a variety of internet devices, including edge servers, set-top boxes, personal computers, and other devices.

First, the simplest way to cache videos is to cache the videos that are accessed most frequently, which can also be referred to as the most popular videos. However, in the real world, how to select the most popular videos faces at least two challenges: first, the user's interests are affected by various factors such as recommendations, viewer gender, age, etc., and thus, the user's video requests change over time, and the video requested by the user is only a small portion of the most popular videos. Meanwhile, the cache space of each device is limited, so that the devices in the internet need to discard useless videos in time and continuously update the cached videos, and therefore the latest interests of users can be captured in time and higher cache efficiency can be achieved. Secondly, in the internet, the storage and bandwidth capacities of devices are heterogeneous, and the services they provide are also diverse, and therefore, it is not possible to simply copy the cache of one device to all other devices.

In recent years, video caching methods for predicting the future popularity of videos by using video history request records are more. This has the advantage that the historical popularity of a video can be obtained simply by calculating the frequency with which the video is requested. However, predicting video popularity needs to rely on a large number of user requests to record, i.e. we cannot accurately predict the popularity of video on devices serving only one or two users. Thus, popularity-based caching algorithms are useless for edge devices, which cannot capture diverse user interests and diverse videos. The point process model can capture the clicking action of the user, the clicking action is expressed in a mathematical form, the self-excited Hox process can predict the future clicking action of the user, the prediction list can be enriched through the mutual excited Hox process, and the prediction is more accurate. By utilizing the Hox process, the future video watching action of each device can be effectively predicted, so that the effects of caching in advance and reducing network flow are achieved.

Disclosure of Invention

In order to solve the defect that the popularity of videos on equipment served by a plurality of users cannot be accurately predicted by using a video history request record in the prior art, the invention provides a video pre-caching method based on a Hox process and matrix decomposition.

In order to solve the technical problems, the technical scheme of the invention is as follows:

a video pre-caching method based on a Hox process and matrix decomposition comprises the following steps:

s1, expressing the proportion of cached requests in all the requests in a period of time by using a hit rate, and mathematically defining the hit rate again by using a mode suitable for a point process according to a hit rate model;

s2, predicting the intensity of all videos and sequencing the videos according to the self history of the equipment by using a self-excited Hox process caching model, and caching the videos with a certain size;

s3, adding influences of neighbor historical clicks on equipment by a cache model in the process of mutually exciting the Hox, predicting the strength of all videos, and improving the hit rate;

s4, carrying out decomposition and dimension reduction on the parameters of the S3 by using a matrix decomposition and dimension reduction model;

and S5, optimizing the mutual excitation Hox process cache model by using a process optimization algorithm, and predicting the future behavior intensity.

In a preferred embodiment, the step S1 includes the following steps:

s11, use

To represent the strength in the hokes process, where u represents the caching device, i represents the video, and t represents the time at which the strength is;

s12, defining the hit rate as follows:

wherein s is_iIs the size of the cached video, B_uIs the buffer space of the device u,

is an indication function indicating whether at time t, device u will buffer video i,

indicating function

Is defined as follows:

s13, the hit rate model, namely the objective function, is expressed as:

in a preferred embodiment, the specific steps at S2 are as follows:

s21, defining epsilon^T＝{t₁,t₂,…,t_KRepresents the click time of all click events, and the recorded time window is]0, T) and has T₁＜t₂＜…＜t_KK represents the total number of events;

s22, define

Then

Wherein b is_uiRepresents the reference or immigration strength of the current process, i.e. the strength value of the process at which no historical events occur, and phi (t-t') represents the activation function of the mutually excited hokes process, i.e.:

φ_ui(t-t′)＝α_uig(t-t′)#(2-4)

α_ui>0 represents how much the current process will be affected by historical events, and for g (t-t'):

g(t-t′)＝exp(-δ(t-t′))#(2-5)

δ >0 is a hyper-parameter representing the rate of decay of an activation or decay function, the greater the value, the faster the decay;

s23, in order to maximize the hit rate defined above, optimizing each process by using an optimization method of the Hox process, wherein a likelihood function is defined as:

after the likelihood function is available, it is derived and then each parameter is estimated using a gradient descent method.

In a preferred embodiment, the specific steps of S3 are as follows:

s31, adding the influence of neighbor historical electrolysis on the current equipment, wherein the influence is defined as follows:

s32, combining (2-3) and (2-7) to obtain a click sequence which considers the historical click event sequence of the click event sequence and the historical click sequences of other devices

The pre-estimated formula of (2):

wherein by delta₁And delta₂The two attenuation functions are defined to represent different degrees of attenuation, respectively, so that the attenuation functions are represented as follows:

g₁(t-t′)＝exp(-δ₁(t-t′))#(2-9)

g₂(t-t′)＝exp(-δ₂(t-t′))#(2-10)

s33, define SE and ME

The impact of each device on its own click process is different from that of other device click processes, so a parameter β e (0,1) needs to be added to balance SE and ME, and then,

is represented as follows:

in a preferred embodiment, the specific steps of S4 are as follows:

s41, performing dimension reduction on the formula (2-13), referring to an SVD algorithm, and performing the step b_uiAnd α_uiThe following approximations are made, respectively:

b_ui＝b_u+b_i#(2-14)

wherein, b_u,b_iRepresenting the basic intensity (immigration intensity) of the device and video, respectively, pair α_uiDoing matrix decomposition, q_i,p_uThe video implicit vector α, the implicit vector of the device, is defined to have d dimensions, so the number of parameters is reduced from 2mn to n + m + nd + md, and m, n > d.

S42, defining other equipment sets influencing equipment u as R_uImprovement of

ME moiety of (1) and finally

The estimated expression is as follows:

so far, the expression has stronger fitting and generalization capability and wider meaning

Equation (2-16).

In a preferred embodiment, the step S5 includes the following steps;

s51, solving parameters according to the pre-estimation formula obtained in the step S4 to maximize an equivalent log-likelihood function:

wherein, the formula (2-17) is a log-likelihood function of the click process of a single device-video, and for all devices and videos, the log-likelihood function is as follows:

maximize equation (2-18), equivalent to minimize-L, to prevent overfitting, increase generalization ability, pair b_u,b_i,q_i,p_uRespectively adding quadratic regular terms to obtain a final optimization target:

s.t.b_u,b_i,q_i,p_u>0,

β_u∈(0,1),

s52, all parameters in the formulas (2-19) are required to be greater than 0 to prevent

Leading to illegal logarithmic calculation, substituting (2-16) into (2-18) yields:

s53, performing partial derivation on the (2-20) to obtain a formula of each parameter under gradient updating:

s54, solving by using a gradient descent method according to the respective gradient formula of each parameter, and assuming that one parameter is theta and the initial value is theta₀Then, the update formula is:

different initial values are set according to specific scenes or data, and then a gradient descent method is used for obtaining suboptimal solutions of all parameters.

Compared with the prior art, the technical scheme of the invention has the beneficial effects that:

the method and the device can effectively predict the watching intensity of each device to different videos in the future, and can remarkably reduce the flow in the future network and improve the user experience by pre-caching some videos with the maximum intensity.

Drawings

FIG. 1 is a flow chart of the steps of the present invention.

Detailed Description

The drawings are for illustrative purposes only and are not to be construed as limiting the patent;

for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;

it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.

The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.

Example 1

As shown in fig. 1, a video pre-caching method based on the hokes process and matrix decomposition includes the following steps:

s1, expressing the proportion of cached requests in all the requests in a period of time by using a hit rate, and mathematically defining the hit rate again by using a mode suitable for a point process according to a hit rate model;

s2, predicting the intensity of all videos and sequencing the videos according to the self history of the equipment by using a self-excited Hox process caching model, and caching the videos with a certain size;

s3, adding influences of neighbor historical clicks on equipment by a cache model in the process of mutually exciting the Hox, predicting the strength of all videos, and improving the hit rate;

s4, carrying out decomposition and dimension reduction on the parameters of the S3 by using a matrix decomposition and dimension reduction model;

and S5, optimizing the mutual excitation Hox process cache model by using a process optimization algorithm, and predicting the future behavior intensity.

Example 2

The video pre-caching method based on the hokes process and the matrix decomposition provided by the embodiment is consistent with the method provided by the foregoing, and only the respective steps are further limited.

A video pre-caching method based on a Hox process and matrix decomposition comprises the following steps:

s1, expressing the proportion of cached requests in all the requests in a period of time by using a hit rate, and mathematically defining the hit rate again by using a mode suitable for a point process according to a hit rate model;

s2, predicting the intensity of all videos and sequencing the videos according to the self history of the equipment by using a self-excited Hox process caching model, and caching the videos with a certain size;

s3, adding influences of neighbor historical clicks on equipment by a cache model in the process of mutually exciting the Hox, predicting the strength of all videos, and improving the hit rate;

s4, carrying out decomposition and dimension reduction on the parameters of the S3 by using a matrix decomposition and dimension reduction model;

and S5, optimizing the mutual excitation Hox process cache model by using a process optimization algorithm, and predicting the future behavior intensity.

In a preferred embodiment, the step S1 includes the following steps:

s11, use

To represent the strength in the hokes process, where u represents the caching device, i represents the video, and t represents the time at which the strength is;

s12, defining the hit rate as follows:

wherein s is_iIs the size of the cached video, B_uIs the buffer space of the device u,

is an indication function indicating whether at time t, device u will buffer video i,

indicating function

Is defined as follows:

s13, the hit rate model, namely the objective function, is expressed as:

in a preferred embodiment, the specific steps at S2 are as follows:

s21, defining epsilon^T＝{t₁,t₂,…,t_KRepresents the click times of all click events, the recorded time window is 0, T), and there is T₁＜t₂＜…＜t_KK represents the total number of events;

s22, define

Then

Wherein b is_uiRepresents the reference or immigration strength of the current process, i.e. the strength value of the process at which no historical events occur, and phi (t-t') represents the activation function of the mutually excited hokes process, i.e.:

φ_ui(t-t′)＝α_uig(t-t′)#(2-4)

α_ui>0 denotes how much the current process will be affected by historical events, for g (t-t') Comprises the following steps:

g(t-t′)＝exp(-δ(t-t′))#(2-5)

δ >0 is a hyper-parameter representing the rate of decay of an activation or decay function, the greater the value, the faster the decay;

s23, in order to maximize the hit rate defined above, optimizing each process by using an optimization method of the Hox process, wherein a likelihood function is defined as:

after the likelihood function is available, it is derived and then each parameter is estimated using a gradient descent method.

In a preferred embodiment, the specific steps of S3 are as follows:

s31, adding the influence of neighbor historical electrolysis on the current equipment, wherein the influence is defined as follows:

s32, combining (2-3) and (2-7) to obtain a click sequence which considers the historical click event sequence of the click event sequence and the historical click sequences of other devices

The pre-estimated formula of (2):

wherein by delta₁And delta₂The two attenuation functions are defined to represent different degrees of attenuation, respectively, so that the attenuation functions are represented as follows:

g₁(t-t′)＝exp(-δ₁(t-t′))#(2-9)

g₂(t-t′)＝exp(-δ₂(t-t′))#(2-10)

s33, define SE and ME

The impact of each device on its own click process is different from that of other device click processes, so a parameter β e (0,1) needs to be added to balance SE and ME, and then,

is represented as follows:

in a preferred embodiment, the specific steps of S4 are as follows:

s41, performing dimension reduction on the formula (2-13), referring to an SVD algorithm, and performing the step b_uiAnd α_uiThe following approximations are made, respectively:

b_ui＝b_u+b_i#(2-14)

wherein, b_u,b_iRepresenting the basic intensity (immigration intensity) of the device and video, respectively, pair α_uiDoing matrix decomposition, q_i,p_uThe video implicit vector α, the implicit vector of the device, is defined to have d dimensions, so the number of parameters is reduced from 2mn to n + m + nd + md, and m, n > d.

S42, defining other equipment sets influencing equipment u as R_uImprovement of

ME moiety of (1) and finally

The estimated expression is as follows:

so far, the expression has stronger fitting and generalization capability and wider meaning

Equation (2-16).

In a preferred embodiment, the step S5 includes the following steps;

s51, solving parameters according to the pre-estimation formula obtained in the step S4 to maximize an equivalent log-likelihood function:

wherein, the formula (2-17) is a log-likelihood function of the click process of a single device-video, and for all devices and videos, the log-likelihood function is as follows:

maximize equation (2-18), equivalent to minimize-L, to prevent overfitting, increase generalization ability, pair b_u,b_i,q_i,p_uRespectively adding quadratic regular terms to obtain a final optimization target:

s.t.b_u,b_i,q_i,p_u>0,

β_u∈(0,1),

s52. requirement publicationAll parameters in the formulae (2-19) are greater than 0 in order to prevent

Leading to illegal logarithmic calculation, substituting (2-16) into (2-18) yields:

s53, performing partial derivation on the (2-20) to obtain a formula of each parameter under gradient updating:

s54, solving by using a gradient descent method according to the respective gradient formula of each parameter, and assuming that one parameter is theta and the initial value is theta₀Then, the update formula is:

different initial values are set according to specific scenes or data, and then a gradient descent method is used for obtaining suboptimal solutions of all parameters.

The terms describing positional relationships in the drawings are for illustrative purposes only and are not to be construed as limiting the patent;

it should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.

Claims (6)

1. A video pre-caching method based on a Hox process and matrix decomposition is characterized by comprising the following steps:

s1, expressing the proportion of cached requests in all the requests in a period of time by using a hit rate, and mathematically defining the hit rate again by using a mode suitable for a point process according to a hit rate model;

s2, predicting the intensity of all videos and sequencing the videos according to the self history of the equipment by using a self-excited Hox process caching model, and caching the videos with a certain size;

s3, adding influences of neighbor historical clicks on equipment by a cache model in the process of mutually exciting the Hox, predicting the strength of all videos, and improving the hit rate;

s4, carrying out decomposition and dimension reduction on the parameters of the S3 by using a matrix decomposition and dimension reduction model;

and S5, optimizing the mutual excitation Hox process cache model by using a process optimization algorithm, and predicting the future behavior intensity.

2. The method for video pre-buffering based on hokes process and matrix decomposition as claimed in claim 1, wherein said S1 comprises the following steps:

s11, use

To represent the strength in the hokes process, where u represents the caching device, i represents the video, and t represents the time at which the strength is;

s12, defining the hit rate as follows:

wherein s is_iIs the size of the cached video, B_uIs the buffer space of the device u,

is an indication function indicating whether at time t, device u will buffer video i,

indicating function

Is defined as follows:

s13, the hit rate model, namely the objective function, is expressed as:

3. the video pre-buffering method based on the hokes process and the matrix decomposition as claimed in claim 1, wherein the specific steps at S2 are as follows:

s21, defining epsilon^T＝{t₁，t₂，...，t_KRepresents the click times of all click events, the recorded time window is 0, T), and there is T₁＜t₂＜…＜t_KK represents the total number of events;

s22, define

Then

Wherein b is_uiA reference or immigration strength, which represents the current process, i.e. the strength at which no historical events of the process occur, phi (t)-t') represents the activation function of the mutual hokes process, namely:

φ_ui(t-t′)＝α_uig(t-t′)#(2-4)

α_ui>0 indicates how much the current process will be affected by historical events, for g (t-t'):

g(t-t′)＝exp(-δ(t-t′))#(2-5)

δ >0 is a hyper-parameter, representing the decay rate of the activation or decay function, the larger the value, the faster the decay;

s23, in order to maximize the hit rate defined above, optimizing each process by using an optimization method of the Hox process, wherein a likelihood function is defined as:

after the likelihood function is available, it is derived and then each parameter is estimated using a gradient descent method.

4. The method for video pre-buffering based on hokes process and matrix decomposition as claimed in claim 3, wherein the specific steps of S3 are as follows:

s31, adding the influence of neighbor historical electrolysis on the current equipment, wherein the influence is defined as follows:

s32, combining (2-3) and (2-7) to obtain a click sequence which considers the historical click event sequence of the click event sequence and the historical click sequences of other devices

The pre-estimated formula of (2):

wherein is made ofδ₁And delta₂The two attenuation functions are defined to represent different degrees of attenuation, respectively, so that the attenuation functions are represented as follows:

g₁(t-t′)＝exp(-δ₁(t-t′))#(2-9)

g₂(t-t′)＝exp(-δ₂(t-t′))#(2-10)

s33, define SE and ME

The impact of each device on its own click process is different from that of other device click processes, so a parameter β e (0,1) needs to be added to balance SE and ME, and then,

is represented as follows:

5. the method for video pre-buffering based on the hokes process and the matrix decomposition as claimed in claim 4, wherein the specific steps of S4 are as follows:

s41, performing dimension reduction on the formula (2-13), referring to an SVD algorithm, and performing the step b_uiAnd α_uiThe following approximations are made, respectively:

b_ui＝b_u+b_i#(2-14)

wherein, b_u，b_iRepresenting the basic intensity (immigration intensity) of the device and video, respectively, pair α_uiDoing matrix decomposition, q_i，p_uThe video implicit vector α, the implicit vector of the device, is defined to have d dimensions, so the number of parameters is reduced from 2mn to n + m + nd + md, and m, n > d.

S42, defining other equipment sets influencing equipment u as R_uImprovement of

ME moiety of (1) and finally

The estimated expression is as follows:

so far, the expression has stronger fitting and generalization capability and wider meaning

Equation (2-16).

6. The method for video pre-buffering based on the hokes process and the matrix factorization of claim 5, wherein said S5 comprises the following steps;

s51, solving parameters according to the pre-estimation formula obtained in the step S4 to maximize an equivalent log-likelihood function:

wherein, the formula (2-17) is a log-likelihood function of the click process of a single device-video, and for all devices and videos, the log-likelihood function is as follows:

for equations (2-18)Row maximization, equivalent to minimizing-L, to prevent overfitting, increase generalization ability, for b_u，b_i，q_i，p_uRespectively adding quadratic regular terms to obtain a final optimization target:

s52, all parameters in the formulas (2-19) are required to be greater than 0 to prevent

Leading to illegal logarithmic calculation, substituting (2-16) into (2-18) yields:

s53, performing partial derivation on the (2-20) to obtain a formula of each parameter under gradient updating:

s54, solving by using a gradient descent method according to the respective gradient formula of each parameter, and assuming that one parameter is theta and the initial value is theta₀Then, the update formula is:

different initial values are set according to specific scenes or data, and then a gradient descent method is used for obtaining suboptimal solutions of all parameters.

Images