WO2015106557A1

WO2015106557A1 - Correction method for quality of experience (qoe) of mobile streaming media user, and server

Info

Publication number: WO2015106557A1
Application number: PCT/CN2014/083088
Authority: WO
Inventors: 陈坚; 吴文峰; 王德政; 申山宏; 程少飞; 周晶; 刘智江; 由李艳; 周文安; 赵立; 华孟
Original assignee: 中兴通讯股份有限公司
Priority date: 2014-01-17
Filing date: 2014-07-25
Publication date: 2015-07-23
Also published as: CN104796443B; CN104796443A

Abstract

Disclosed are a correction method for the quality of experience (QoE) of a mobile streaming media user, and a server. The method comprises: receiving user behaviour data and current video damage data; according to the user behaviour data, obtaining a value of influence on the QoE of an individual; according to the current video damage data, obtaining initial user quality of experience QoE_init; and according to the value of influence on the QoE of the individual and the initial user quality of experience QoE_init, obtaining corrected individual user quality of experience QoE_final.

Description

Mobile streaming media user experience quality ΟθΕ correction method and server

The present invention relates to the evaluation and management of user experience quality in the field of wireless network communication, and more particularly to a method for QoE correction of mobile streaming media user experience quality.

BACKGROUND With the continuous development of mobile communication technologies, the demand for mobile streaming media services has exploded, and the importance of performance evaluation has become increasingly prominent. Quality of Experience (QoE) is a concept of end-to-end user. It measures the pros and cons of the service from the perspective of the user. It is a comprehensive subjective experience of the user on the current network, service quality and performance. The development of HTTP Streaming technology is to improve the user experience. Therefore, Quality of Experience (QoE) is the core indicator for measuring the performance of HTTP streaming services. The streaming media service based on the HTTP protocol is different from the traditional streaming media service based on the UDP protocol. Because the underlying network uses a reliable TCP protocol, the reason for the degradation of video quality is that the video caused by the retransmission of the packet is too late to fill the empty buffer. Therefore, the network layer factor that affects the performance of the video service based on the HTTP protocol is mainly the delay. , packet loss and network bandwidth, application layer factors are mainly time factors such as initial buffering and re-buffering. At present, there are two methods to evaluate the quality of user experience in mobile streaming media services: (1) Subjective quality evaluation, that is, the participants in the test score the damaged video clips tested, and subjectively evaluate the video quality to evaluate The quality of the business, typically including Single Stimulus Continuous Quality Evaluation (SSCQE), the Double-Stimulus Continuous Quality-Scale method (DSCQS), and the double-stimulus damage standard The Double Stimulus Impairment Scale method (DSIS). Subjective evaluation is highly accurate because it directly reflects the subjective feelings of people, but this evaluation method is not suitable for real-time evaluation, and this method requires a lot of manpower and material resources, and the workload is huge and time-consuming, and it is generally difficult to implement. (2) Objective quality evaluation, that is, establishing a mathematical model for video data, and obtaining parameters reflecting the video quality through a series of calculations, and obtaining the final evaluation result, which is divided into full reference according to the degree of reference to the original reference video (Full Reference, FR) , Reduced Reference (RR) and No Reference (NR) three video quality evaluation methods. Common reference signal to noise ratio (PSNR), Mean Square Error (MSE) and other full reference evaluations and some reference evaluations need to refer to all or part of the source video, so it is not applicable to the network. HTTP streaming service in transit. In contrast, the no-reference evaluation method does not need to refer to the source video, and only evaluates according to the characteristics of the damaged video itself, so it has become the focus of academic research. However, the current non-reference evaluation method is still not perfect. How to effectively map the quantifiable and measurable QoS parameters with QoE needs further study. In addition, the quality of user experience depends not only on the network layer, but also on the application layer, user layer, and business layer. Therefore, the simple objective evaluation method does not reflect the subjective feelings of people. , can't be close to the user's perception. How to effectively combine subjective evaluation with objective evaluation still needs further research. In related patents and related papers, most people directly map QoS and QoE through the network layer, but this only considers the network factor to ignore the impact of other layers on QoE; some cache through the combination of network layer and application layer Prediction of zone status, this method can well predict whether QoE is about to be reduced but cannot directly predict the specific QoE value; Some consider the impact of user layer user operation behavior on the buffer area, but does not take into account the user's behavioral response. The user psychology is out, and the user layer parameters cannot be applied to get QoE. According to the above analysis, the relationship between the network layer, the application layer, and the user layer at the third layer and QoE is as shown in FIG. 1 . Then on this basis, further analysis of how to use the user layer information for QoE evaluation. The quality of user experience is more subjective and individualized, and it is influenced by many factors. We can see that the existing evaluations only consider some factors affecting QoE, some are only a single QoS and QoE mapping, and some directly simplify QoE into a buffer state. While the user is watching the movie, there will be a series of operational behaviors such as pause, change resolution and exit. Some of these behaviors may have some impact on the quality of the streaming media service, and some may express some psychological state of the user. The evaluation methods of related technologies tend to idealize the scenario, assuming that the network conditions are unchanged, and ignoring the impact of the user's operation behavior in the process of watching the video on the quality of the streaming media service. This has a large deviation from the actual use environment of the user, so ultimately Evaluation results are easy with the user's actual body The inspection quality is inconsistent.

SUMMARY OF THE INVENTION The embodiments of the present invention are directed to a large deviation of QoE evaluation in a mobile streaming media service based on the HTTP protocol, and a mobile streaming user quality QoE correction method and server are proposed to correct a user personalized QoE evaluation. A mobile streaming media user experience quality QoE correction method, the method comprising: receiving user behavior data and current video impairment data; obtaining an impact value on the personal QoE according to the user behavior data; and obtaining an initial user according to the current video impairment data quality of experience QoE _MIT; corrected individual users according to the quality of experience QoE _FMAl influence values of the individual and the initial QoE quality of experience QoE _mit. Preferably, the method further includes: the impact value of the personal QoE includes: a first influence value Et of the fluency on the individual QoE and a second influence value Ep of the clarity on the individual QoE; The step of obtaining the influence value of the user behavior data on the personal QoE includes: establishing a user behavior table according to the user behavior data, and calculating the E _t and E _p according to the user behavior table. Preferably, the method further includes: the user behavior table includes:

L _qtl ,, L _qtr, Lq _fr denote a user viewing a long history of the average degree of initial buffering time of Q-based video quit, then the average level of the buffer length and the average degree of frequency re-buffering;

N _pause , indicating the total number of passive pauses for this video user; and Time, representing the length of the played video; the steps of calculating the E _t and E _p according to the user behavior table include: the E _t =<uID, {(C ₁ , E _lt ); (C ₂ , E _2t ); (C _n , E _n t)}>; where, E _{ltu ≤ 1 ≤ n} , which indicates the influence of the Q-type video fluency on the personal QoE of the user, l < E _lt <0;

Where I _lt =I(Lq _tl , Lq _tr , L _qfr ) represents the influence of the user's passive exit behavior on QoE; M(L _pause ) represents the influence of the user's passive pause behavior on QoE; _ei + _e2 = l, ei , e ₂ respectively represents the passive exit behavior of the user and the influence coefficient of the passive pause behavior on the QoE of the viewing video; when the user is cold-started, ^ 2 ⁰ , then E _it = M (L)

I _it = - 1 + (inLqti + uiLqfr + mLqtr) 13 ( _i < i. _t <0 ) . where i 〃 + to ≤ l;

M(L _pa , _IS e) = — 1 , (-1 <M(L _pause )<0) ,

_JL in T = ^ ^pause ·

Time E _p =<uID,{(C ₁ ,E _lp );(C ₂ ,E _2p );...(C _n ,E _np )}>; where, E _{ip (1≤1≤n)} Indicates the impact of a certain type of video Q resolution on the user's personal QoE:

■1<E _1P <1;

Ei _P = mlip + ; (Npic); where I _ip =I(F _pic — _m ) represents the user's influence on the QoE of the historical expectation of video clarity; M(Np _K ;) indicates the clarity of the video The influence of user emotion on QoE; 1^+112=1,1^, n ₂ respectively represent the historical expectation of the user's definition of the video and the influence coefficient of the current emotion on the individual QoE; when the user is cold start, Lip = 0 , at this time Ε _Ψ = niM(N _pi c); H , ( -i<i _iP ≤o ) ; where F . . = ^N «i , N _m . _{Ln C} ,. indicates that when viewing the video of the ^ category, the user has performed

_ Jy

Large resolution "the number of videos behaved; N. indicates the total number of videos that the user viewed (^ category; k<0;

M(Npic) = w(N _P ic _ in - Npic _ de) , j <Μ(Ν■ )< 1 ) · where N _PIE — _m represents the number of times the resolution is increased during the viewing of the video; N _PIE _ _DE indicates the number of times the resolution was lowered during the video viewing. Preferably, the method further includes: the current video damage data comprises: an initial buffer duration T _mi , a rebuffer duration T _rebuf and a rebuffer frequency F _rebuf ; the rebuffer duration T _rebuf is a video automatic pause Or the re-buffer duration when automatically exiting; if the video buffer can maintain the video to continue playing, it is judged that the pause or exit is automatic pause or automatic exit; the step of obtaining the initial user experience quality QoE _mit according to the current video damage data includes: The initial user experience quality QoE _{mit is} obtained according to the reception of the initial buffer duration T _im , the rebuffer duration T _rebuf and the rebuffer frequency F _rebuf . Preferably, the method further comprising: the value of the impact of the individual and the initial QoE Quality of Experience QoE _{mi The} user correction to user Quality of Experience QoE _FMAl step comprises: leg _F QoE obtained according to the formula _T _1:

QoEfmal = QoEinit + wixEt + miEp . where -l≤E _t ≤0 , -1<Ερ<1 , m\ + m2 = l , ml and m2 are the weight coefficients of E _t and E _p , respectively. A mobile streaming media user experience quality QoE modified server, the server comprising: a data receiving module, configured to receive user behavior data and current video impairment data; and a user behavior recording module, configured to obtain an impact value on the personal QoE according to the user behavior data;

The QoE initial evaluation module is configured to obtain an initial user experience quality QoE _mit according to the current video damage data;

The QoE correction module is configured to obtain a modified user experience quality QoE _fmal according to the impact value of the personal QoE and the initial user experience quality QoE _mit . Preferably, the server further has the following characteristics: the impact value of the personal QoE includes: a first influence value Et of the fluency on the personal QoE and a second influence value Ep of the clarity on the individual QoE; The user behavior recording module is configured to obtain an impact value on the personal QoE according to the user behavior data in the following manner: establishing a user behavior table according to the user behavior data, and calculating the E according to the user behavior table _t and E _p . Preferably, the server further has the following features: The user behavior table includes:

_{_{_{L qtl ,, L qtr, L qfr}}} , respectively, a user viewing history (^ length extent an average initial buffering video class exit, then the average degree of the buffer length and the average degree of frequency re-buffering;

N _pause , indicating the total number of times the video user passively pauses; and time, indicating the length of the played video; the user behavior recording module is configured to calculate the E _t and E according to the user behavior table in the following manner _p : the E _t =<uID,

... (C _n ,E _nt )}>; Where E _lt(1≤1≤n ) represents the influence of a certain type of video C on the personal QoE of the user, l<E _lt <0;

E _it =eiI _it + e ₂ M(L pause where I _lt =I(Lq _tl , Lq _tr , L _qfr ) represents the influence of the user's passive exit behavior on QoE; M(L _pause ) represents the user's passive pause behavior pair The influence value of QoE; _{ei + e2} = l, ei, e ₂ respectively represent the passive exit behavior of the user and the influence coefficient of the passive pause behavior on the QoE of the watched video; when the user starts coldly, ^{= υ} , then E _it = e ₂ M (L pause

I _it = -1 + (UlLqti + UlLqfr + XnLqtr) I 3 ( \<[ < 0 ) . where Lq _tl , Lq _tr , L _qfr represent the average initial buffer duration when the user history watches the Q video exits, respectively. , the average degree of re-buffering and the average degree of re-buffering frequency; + +to ; M ( earning ) = e- -- 1 , ( - 1 < M ( L _pause ) < 0 ) , Lizhong _ ^ pause .

Eight ' pause .■ ,

Time E _p =<uID,{(C ₁ ,E _lp );(C ₂ ,E _2p );...(C _n ,E _np )}>; where Ε _{ιρ( 1≤1≤n)} Indicates the influence of the d-type video definition on the user's personal QoE, -1<Ε _ιρ <1; Ei _P = ndip + niM (N _P w); where I _ip =I(F _pic — _m ) indicates the user-to-video The history of sharpness expects the value of its influence on QoE; M(N _pic ) represents the influence of the user's emotion on the QoE caused by the clarity of the video; n +nf in ₂ respectively represents the historical expectation of the user's definition of the video and The coefficient of influence of the current mood on the individual QoE; when the user is cold-started, /* = 0, at this time = (Λ^.);

I _ip F _{pic m} , ( -ΐ<ι _ιρ <ο ) ; its, medium ₌ ^{N —} m— _Ci

1 *■ pic m _Τ pic— m— Indicates that the user has performed the video when viewing the video of the ^ category

iV

Large resolution "the number of videos behaved; N. indicates the total number of videos that the user viewed (^ category; k<0; M(Npic) = w(Npic― in - Npic— A) , (_ i <M(Npi _c )< 1 ); where N _pic — _m represents the number of times the resolution is increased during the viewing of the video; N _pic — _de indicates the number of times the resolution was lowered during the video viewing. Preferably, the server further has the following features: The current video damage data includes: an initial buffer duration T _mi , a rebuffer duration! ^^ and rebuffer frequency? ^^ The re-buffer duration T _rebuf is the re-buffer duration when the video is automatically paused or automatically exited; if the video buffer can maintain the video to continue playing, it is judged that the pause or exit is automatic pause or automatic exit; the QoE initial The evaluation module is configured to obtain an initial user experience quality QoE _mit according to the current video damage data in the following manner: receiving the initial buffer duration T _mi , the rebuffer duration T _rebuf , and the rebuffer frequency F _{rebuf to} obtain an initial user experience Quality QoE _imt . Preferably, the server further has the following features: the QoE correction module is configured to correct the user experience quality QoE _fmal according to the impact value of the personal QoE and the initial user experience quality QoE _mit in the following manner: The E _t , E _p and QoE _mit get QoE _fmal:

QoE final = QoEinit + JfllEt + TUlEp . where -l ≤ E _t ≤ 0, -l < Ep < l , ml + m2 = l , ml, m2 are the weight coefficients of E _t and E _p , respectively. The embodiment of the present invention further provides a computer program, including program instructions, when the program instruction is executed by a server, so that the server can execute the method of the foregoing embodiment.

Embodiments of the present invention also provide a carrier carrying the above computer program. In summary, the method and the server according to the embodiment of the present invention have the following beneficial effects:

1. The embodiment of the present invention can obtain user information in a timely and effective manner by performing data collection and analysis on the terminal and the server, and has little impact on the user.

The historical expectation of the video and the emotions when watching the video are compared with the existing methods of obtaining user feedback through questionnaires, and the user's psychology can be obtained without a lot of manpower and resources, and is simple and easy to implement.

3. The embodiment of the present invention fully considers the user's expectation of the video on the subjective level of the user and the influence of the user's emotion on the user's personal experience quality, and overcomes the drawbacks of the existing method that the user does not consider the subjective feeling of the user; Subjective level video expectations and user sentiment reduce the popularity of QoE evaluation, making the resulting QoE more accurate to the user.

4. The embodiment of the present invention fully considers different requirements of different users for video clarity, and overcomes the drawbacks of considering the video fluency and ignoring the picture quality in the existing HTTP streaming media evaluation method, so that the user experience evaluation is closer to the user perception. 5. The embodiment of the present invention can be combined with network QoS. The operator can classify users according to the user's personalized QoE evaluation, and perform more reasonable network resource allocation, so as to avoid the situation that the network parameters are good and the user experience is poor.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram showing the relationship between layers in the embodiment of the present invention; FIG. 2 is a schematic diagram showing the QoE modification of the mobile streaming user experience quality according to an embodiment of the present invention; FIG. 4 is a structural diagram of a mobile streaming user experience quality QoE correction system according to an embodiment of the present invention. BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that, in the case of no conflict, the features in the embodiments and the embodiments in the present application may be arbitrarily combined with each other.

The problem of current HTTP mobile streaming user experience quality evaluation: (1) Subjective evaluation method is time-consuming and laborious to measure data, and cannot be feedback in real time, and users are not necessarily willing to participate actively, which causes problems in data collection; (2) Mapping through network QoS QoE evaluation method, no user participation, can not accurately and timely reflect the user experience; (3) the existing user experience quality evaluation does not distinguish the user to the individual, easily cause unnecessary waste of network resources and error evaluation of individual users QoE .

As shown in FIG. 2, an embodiment of the present invention provides a schematic diagram of a QoE modification method for a mobile streaming user experience quality. The embodiment of the present invention directly records the damage degree of the video according to the trigger of the video damage, including the initial buffer duration of the video, Re-buffer duration and re-buffering frequency, and then get the initial QoE; then through the user to watch the historical behavior record of the same type of video and the current viewing behavior record to predict the impact of video fluency and video definition on the user's personal QoE, and then The initial QoE is corrected to obtain the final QoE value for the user. In the embodiment of the present invention, user habits, user expectations, and user emotions are reflected according to user layer parameters, and then QoEs obtained according to user habits, user expectations, and user emotion corrections are corrected to obtain a final QoE value for the user. However, in the embodiment of the present invention, the QoE is modified to directly use the user layer parameters. User layer parameters, preferably, may refer to current and historical viewing behaviors, including but not limited to pause, exit, and change resolution.

As shown in FIG. 3, an embodiment of the present invention provides a mobile streaming media user experience quality QoE correction method, which creatively proposes a method for responding to different users' expectations of different types of videos, and subdivides the user's expectations into users. The expectation of fluency and the desire for clarity, thereby correcting the user's overall satisfaction with the video by analyzing the operational behavior of the user during video viewing, thereby achieving the purpose of the user to personally correct the user experience quality evaluation. A mobile streaming user quality of experience QoE correction method, comprising: Step 101: Receive user behavior data and current video impairment data. User behavior data can reflect the clarity and fluency of the video; and user behavior data can also reflect the user's emotions, habits, expectations, and so on. Step 102: Obtain an impact value on the individual QoE according to the user behavior data; the impact value on the individual QoE includes: the first influence value Et of the fluency on the individual QoE and the second influence value Ep of the clarity on the individual QoE; according to the user behavior The steps of obtaining the influence value of the data on the personal QoE include: establishing a user behavior table according to the user behavior data, and calculating E _t and E _p according to the user behavior table. The user behavior table includes: uID, Q, user-act, T _mi , T _rebuf , F _rebuf ; where uID is the unique identifier of the user;

... , C _n } is the video category viewed by the user; user_act={ "pause", "quit", "pic-in", "pic-de"} is the user's operation behavior, and the content included therein respectively represents " Pause ",""exit","increaseresolution" and "reduce resolution"; time records when the video has been played when the action is triggered; T _mi , T _rebuf , F _rebuf indicates the video damage that was present when the action was triggered, The initial buffer duration, rebuffer duration and rebuffer frequency are respectively calculated. According to the user behavior table, E _t and E _p are calculated as:

E _t =<uID,

... (C _n , E _nt )}>; where E _{ltu ≤ 1 ≤ n)} represents the influence of the Q-type video fluency on the user's personal QoE, l < E _lt <0; E _it = l _{it + e2} M (L _p _{_{wherein, I lt = I (L qtl}} , L qtr, L qfr) indicates that the user passive exit influence value against the QoE; M (L _pause) indicates that the user passive suspension affects the value against the QoE; _{ei + E2} = i , ei , e ₂ respectively represent the passive exit behavior of the user and the influence coefficient of the passive pause behavior on the QoE of the watched video; When the user starts cold, ⁰ , then E _it = e ₂ M (L _PAUSE

I _IT = -1 + (UlLqti + UlLqfr + 0 ) .

Where Lq _tl , Lq _tr , L _qfr respectively represent the average initial buffer duration, the average rebuffer duration and the average rebuffer frequency when the user history watches the Q video exit; _WL + W ₂ + W ₃ ≤ l;

M ( e-- - 1 , (-1 <M(L _pause )<0) , where L _ause =^i , N _pause represents the total number of passive pauses of the video user watching this time; time

p ^ause time

Indicates the length of the video being played; v≥0;

E _p =<uID,{(C ₁ ,E _lp );(C ₂ ,E _2p );...(C _n ,E _np )}>, Ε _{ιρ(1≤1≤η} ) indicates that the Q video is clear The influence of degree on the user's personal QoE, -1≤Ε _ιρ ≤1;

Ei _P = mlip + ητΜ (N _P ic); where I _ip =I(F _pic — _m ) represents the user's influence on the QoE of the historical expectation of video sharpness; M(N _pie ) represents the video definition The effect of the user's emotion on QoE; 1^+112=1,1^, n ₂ respectively represent the historical expectation of the user's definition of the video and the influence coefficient of the current emotion on the individual QoE; , hp = 0, at this time Εφ = ηιΜ(Ν _ρΐ α);

H _m ( -i ≤ i _iP <o ); wherein, F. . : ^N - j - ^Ci , N _m . _mCT indicates that the user has performed the video when viewing the video of the category

- 7 V c, — ―

Large resolution "The number of videos that behave; N. indicates the total number of videos that the user viewed (^ category; k<0; M(Npic) = w(Npic- in - Npic-A) , (_ i <M(Npi) _c )< 1 ); where N _pic — _m represents the number of times the resolution is increased during the viewing of the video; N _pic — _de represents the number of times the resolution is reduced during the viewing of the video. Step 103: According to the current video damage data Initial user experience quality QoE _mit ; current video damage data includes: initial buffer duration T _mi , rebuffer duration T _rebuf and rebuffer Frequency F _rebuf ; Re-buffer duration T _rebuf is the re-buffer duration when the video is automatically paused or automatically exited; if the video buffer can maintain the video to continue playing, it is judged that the pause or exit is automatic pause or automatic exit; according to the current video damage initial data obtained quality of experience QoE _mit user comprises: according to the received initial buffer long when T _mi, and then again buffer length _T rebuf buffer to obtain an initial frequency _F rebuf user quality of experience QoE _imt. Step 104: Correct the personal user experience quality QoE _fmal according to the impact value on the individual QoE and the initial user experience quality QoE _mit . The steps of correcting the user experience quality QoE _fmal according to the impact value of the individual QoE and the initial user experience quality QoE _mit include: QoE _fmal according to E _t , E _p and QoE _mit _:

QoE final = QoEinit + JfllEt + TUlEp . where -l ≤ E _t ≤ 0, -l < Ep < l , m\ + m2 = \ , ml, m2 are the weight coefficients of E _t and E _p , respectively.

As shown in FIG. 4, the embodiment of the present invention further provides a server for mobile streaming media user experience quality QoE correction, where the server includes a data receiving module, a user behavior recording module, a QoE initial evaluation module, and a QoE correction module, where: a data receiving module, Set to receive user behavior data and current video impairment data. User behavior data can reflect the clarity and fluency of the video; and user behavior data can also reflect the user's emotions, habits, expectations, and so on. The user behavior recording module is configured to obtain an influence value on the personal QoE according to the user behavior data; The impact on personal QoE includes: fluency on personal QoE first impact value Et and clarity on personal QoE second impact value Ep; user behavior logging module can be set to get personal QoE based on user behavior data in the following manner Impact value: Based on the user behavior data, a user behavior table is established, and E _t and E _p are calculated according to the user behavior table. The user behavior table includes: uID, Q, user-act, T _mi , T _rebuf , F _rebuf ; where uID is the unique identifier of the user;

E _t =<uID,

(C _n , E _nt )}>; where E _{ltu ≤ 1 ≤ n)} represents the influence of the Q-type video fluency on the user's personal QoE, l < E _lt <0;

_{_{_{E it = I it + e2 M}}} (L p _{_{wherein, I lt = I (L qtl}} , L qtr, L qfr) indicates that the user passive exit influence value against the QoE; M (L _pause) represents a user passive suspension behavior QoE The influence value; _{ei + e2} = i , ei , e ₂ respectively represent the passive exit behavior of the user and the influence coefficient of the passive pause behavior on the QoE of the watched video; when the user is cold-started, ⁰ , then E _it = e ₂ M (L _pause ) ;

I _it = -1 + (UlLqti + UlLqfr + 0 ) .

Where Lq _tl , Lq _tr , L _qfr respectively represent the average initial buffer duration, the average rebuffer duration and the average rebuffer frequency of the user history when viewing the Q video exit; _Wl + W ₂ + W ₃ ≤ l ; M(L _P ause) = - 1 , (-1 <M(L _pause )<0), where L ₌ ^i , N _pause represents the total number of passive pauses of the video user watching this time;

Time

Indicates the length of the video being played; v≥0;

E _p =<uID, { (d ,E _lp );(C ₂ ,E _2p ); ... (C _n ,E _np )} > , E _{ip U≤1≤n} ) means class video clarity to the user The influence value of personal QoE, -1≤Ε _ιρ ≤1;

Ei _P = mlip + niM (N _P w); where I _ip =I(F _pic — _in ) represents the user's historical expectation of video sharpness affecting its QoE; M(N _pic ) represents the video definition The resulting impact of user sentiment on QoE;

n ₂ respectively represents the historical expectation of the user's definition of the video and the influence coefficient of the current mood on the individual QoE; when the user is cold-started, lip = 0, then Ε, ρ = ητΜ(Ν _Ρ , ο);

H , (-i ≤ i _iP ≤ o );

Where ₌ N _{Vic m Ci} indicates that when viewing the video of the category ^, the user has performed

¹ pic m Nc pic_m_L7 ^{¥ 1}

M(Npic) = w(Npic― in - Npic— ) , (_ i <M(N _pic )< 1 ); where N _pie — _in represents the number of times the resolution is increased during the viewing of the video; N _pic — _De indicates the number of times the resolution is lowered when watching the video.

The QoE initial evaluation module is set to obtain the initial user experience quality QoE _imt according to the current video damage data; the current video damage data includes: initial buffer duration T _mi , rebuffer duration T _rebuf and rebuffer frequency F _rebuf ; The duration T _rebuf is the re-buffer duration when the video is automatically paused or automatically exited; the video buffer can maintain the video to continue playing, then it is judged that the pause or exit is automatic pause or automatic exit; The QoE initial evaluation module can be configured to obtain an initial user experience quality Q ₀ E _mit according to current video impairment data in the following manner _: receiving initial buffer duration T _mi , rebuffer duration T _rebuf and rebuffer frequency F _{rebuf to} obtain initial user experience Quality QoE _mit . QoE correction module, set to based on the impact on personal QoE and initial user experience quality

QoE _mit gets corrected user experience quality QoE _fmal .

The QoE correction module can be set to correct the user experience quality according to the impact value on the individual QoE and the initial user experience quality QoE _mit in the following way _: QoE _fmal is obtained according to the following formula _: QoE final = QoEinit + JfllEt + miEp . l ≤ E _t ≤ 0, -l < Ep < l , m\ + m2 = \ , ml, m2 are the weight coefficients of E _t and E _p , respectively. For other functions of this server, please refer to the description of the method content. Optionally, the server is divided into a user terminal and a server. The user terminal may include a user behavior monitoring module, a video damage monitoring module, and a data integration sending module: the user behavior monitoring module is responsible for monitoring and recording the user's operating behavior during the viewing process, including pausing, changing the resolution, and exiting; the video damage monitoring module Responsible for collecting information on the application layer, including initial buffer duration, rebuffer duration, and rebuffer frequency; the data integration send module is responsible for integrating the data and sending it to the server. The server side can refer to the description of the above server.

The embodiment of the present invention further provides a computer program, including program instructions, when the program instruction is executed by a server, so that the server can execute the method of the foregoing embodiment.

Embodiments of the present invention also provide a carrier carrying the above computer program. An application example of the present invention is as follows: Step 1: The server side establishes a user behavior table. The user behavior table (uIDA^ser-act, time, T _im , T _rebuf , F _rebuf ) is used to record the operational behavior of the user during the viewing process in order to analyze the user's expectations and current emotions. uID is the unique identifier of the user (can be identified by the terminal number or user ID);

... , C _n } is the video category that the user watches (the classification of the video can be directly classified according to the user's acquisition of the video source, that is, the general video website divides the video into movies, TV dramas, variety shows, entertainment, sports, news, etc.); user_act ={"pause","quit","pic_in","pic_de"} are user action behaviors, including "pause", "exit", "increase resolution" and "reduced resolution"; Record when the video has been played when the action is triggered;

T _mi , T _rebuf , and F _rebuf indicate the existing video impairments when the behavior is triggered, which are the initial buffer duration, the rebuffer duration, and the rebuffer frequency. Step 2: The user starts the streaming media service, and the terminal starts collecting the application layer and the user layer data. (1) The user starts the streaming media service, and triggers the terminal to collect the application layer data. After the video starts playing, the terminal automatically records the initial buffering time of the video; whenever the video is automatically paused (non-human trigger pause button), the terminal marks the occurrence of the re-buffering event, and records the pause time of the video and restarts the playing time. (The time difference is the length of the re-buffering); After each recording, the terminal automatically extracts the previous record, and the frequency of the automatic re-buffering of the statistical video (the number of times of re-buffering/the duration of the played video (s)), the video buffering Average duration.

(2) The user behavior is triggered, and the terminal sends and records the user behavior data into the user behavior table. When the user triggers a certain behavior, the user's identity, the watched video category, the behavior status, the video playing time, and the video damage at this time are recorded in the user behavior table; When the user triggers the "pause" behavior and "exit, the behavior, the terminal automatically recognizes the state of the buffer at this time. If the buffer status is good, the pause behavior is regarded as an active pause due to the user's personal reasons. The active exit is not recorded in the user behavior table. Step 3: The server initially predicts QoE based on the collected application layer data. The server collects the application layer data according to the terminal (the initial buffer duration T _mi during the video watching, rebuffering) The frequency F _rebuf and the re-buffering time T _rebuf ) are divided into three levels: "Low,""Medium," and "High", which are represented by the scores of "1,," 2, and "3," respectively. And then perform the fitting to get:

QoE _mtl = 4.23 - 0.0672L _tl - 0.742L _fr - 0.106L _tr [1]

L _tl , L _fr and indicate the degree of initial buffer duration, the degree of rebuffering frequency, and the degree of rebuffer duration. The recorded video impairments are converted to corresponding degree values as shown in Table 1. Table 1 video damage degree classification

Step 4: Correct the QoE _mit of the initial prediction.

(1) Extract user information from the user behavior record table. Obtaining the characteristics of the user watching the video according to the historical viewing behavior record statistics (including the average initial buffering time when the user watches the video exiting)

_Rebuffer duration T _qrebuf — _C1 , rebuffer frequency F _qre buf_ _C1 ), the number of videos that have undergone the “increase resolution” behavior when viewing this type of video in user history

According to the operation record of the movie viewing, the user increases the resolution, the number of behaviors Np _lc — _m , the number of behaviors of the reduced resolution Np _lc — _de , and the user viewing the video. The number of "pause" behaviors performed during the video is N _pause . When the user starts coldly, he only needs to obtain the operation behavior of the user to watch this movie.

(2) QoE correction for individual users

QoE final = QoEinit + JfllEt + TUlEp where E _t represents the effect of video fluency on the user's personal QoE, and E _p represents the impact of video clarity on the user's personal QoE. -l≤E _t ≤0, -1<Ερ<1 , mi +m ₂ = l , m _x , m ₂ are the weight coefficients of the video fluency influence value and the video sharpness influence value, respectively, by statistical methods (or Other methods such as analytic hierarchy) are obtained. The fluency influence value E _t =<uID,{(C ₁ ,E _lt );(C ₂ ,E _2t );...(C _n ,E _nt )}>, d _u≤1≤n) represents the video Different classifications, E _lt (! < _1≤n , indicates the _influence of a certain type of video d fluency on the user's personal QoE, -l<E _lt <0 , ( "0""-1 " respectively indicates that the user is fluent on the video The expectation of degree is "doesn't care" and "high expectation," the absolute value tends to 1 indicating that the higher the user's expectation of video fluency, the greater the impact on his personal QoE.

E _it = I _it + _e2 M(L _p where I _lt =I(Lq _tl , Lq _tr , L _qfr ) represents the impact value of the user exit behavior on QoE, and the user exit behavior is divided into active exit (user exits for its own reasons) ) and passive exit (the user quits because the quality of the experience is unbearable), the user behavior monitoring module judges the user's exit behavior category according to the state of the cache area when the user exits, and only records the user passively exiting, because the user exits, the service is terminated, so only It is necessary to consider the exit behavior of the user's history when watching such video, that is, to judge the user's tolerance to video damage according to the average video damage degree when the video is passively exited according to the statistical user history, and then obtain the exit behavior for the user to watch this time. The impact of video subjective experience; M (L _pause ) indicates the impact of user pause behavior on QoE, user pause behavior is divided into active pause (the user pauses for their own reasons) and passive pause (the user is suspended due to dissatisfaction with the current experience quality), The user behavior monitoring module determines the user's pause behavior category according to the state of the buffer when the user pauses, and only remembers The user passively pauses, because the user pauses the method used by the user to improve the current video fluency, expresses the user's expectation and mood for the current video quality, so only need to consider the pause behavior when the user currently watches the video, that is, according to the statistical user viewing the current The number of passive pauses performed by the video to determine the extent to which the user expresses dissatisfaction with the video quality, and then infers the effect of the pause behavior on the QoE evaluation of the user watching the video; _ei + _e2 = 1 , and e ₂ represent the user's exit behavior, respectively And the coefficient of influence of the pause behavior on the QoE of this watch video, which can be obtained by statistical methods (or other methods such as analytic hierarchy analysis). When the user starts cold, I _it = 0, then E _it = e ₂ M (L _p

I _it = -1 + (mLqti + U2∑qfr + U?>Lqtr) I 3. ( _1 ≤ 1 ≤ 0 ) where Lq _tl , Lq _tr , L _qfr represent the average initial buffer duration, average rebuffer duration, and average rebuffer frequency of the user history when viewing the Q video exit, respectively. The average initial buffer duration T _qlm — _C1 , the rebuffer duration T _qrebuf — _α , and the rebuffer frequency F _qrebuf — _α obtained by the user from the historical behavior table when the user watches the video exit is the video when the user exits. The degree of damage, "1", "2" and "3" respectively indicate the "low", "medium" and "high" of the degree of damage (the classification is based on the reference table 1). m + U2 + ≤\ , _Ul , u ₂ , u ₃ can be obtained by fitting a large amount of data (or other methods such as analytic hierarchy analysis). When the user history watches the passive exit of this type of video, the lower the damage degree, the lower the acceptable degree of damage to the video. The user can think that the expected value of the video is higher. When the user expects the higher value, the user experience evaluation. The more severe it is, the lower the QoE value of the user under the same damage conditions.

M ( e-— - 1. (-1 <M(L _pause )<0) where, L _ause = ^i ( N _pause represents the total number of times the video user passively pauses, and time represents the length of the video being played.) V≥0, V can be obtained through experimental statistics. The user has paused the operation, which will inevitably have higher expectations before the fluency of the video is less. As the number of pauses increases, the emotion will become more and more dissatisfied, resulting in the same The QoE value of this user's viewing will be reduced under video impairment conditions. The resolution impact value E _p =<uID,{(C ₁ ,E _lp );(C ₂ ,E _2p );...(C _n , E _np )}>, _{( 1 ≤ 1 ≤ n} ) indicates different classifications of video, E _{ip U ≤ 1 ≤ n)} indicates the influence of a certain type of video Q resolution on the user's personal QoE, -1<Ε _ιρ <1 , (The more the absolute value tends to 1, the greater the impact.)

Ei _P = mlip + ητΜ (N _P ic). Where I _ip =I(F _pic — _m ) represents the user's influence on the QoE of the historical definition of the video sharpness, that is, the user is judged according to the frequency of the statistical user increasing the resolution of the video history. The expectation of the definition of the video, and then the user's influence on the subjective experience of the video; M (N _pic ) indicates the impact of the user's emotion on the QoE caused by the clarity of the video, that is, according to the user's video during the viewing Increasing the difference in the number of times of resolution reduction determines the user's satisfaction with the current video definition, and sequentially obtains the influence of the user's viewing video sentiment on the user QoE; 1^+112=1,1^, respectively indicating that the user Historical expectations of video-like clarity and the coefficient of influence of current emotions on individual QoE, It can be obtained by statistical methods (or other methods such as analytic hierarchy analysis). When the user starts cold, there is no history, hp = 0, then Ew = "2 (A ) i _ip = 3⁄4 _em . (-1≤Ι _ιρ ≤0 ) ( "0,,"-Γ, respectively The user's expectation of video clarity is "doesn't care" and "higher expectation". The more the absolute value tends to 1, the higher the user's expectation of video clarity, the greater the impact on his personal QoE.) Among them, F . . : ^N - j - ^Ci ( N _{mc m C} ,. means to watch (^ when the video of the category is played by the user)

7 V c, ~

Large resolution "The number of videos that behave; N. indicates the total number of videos that the user is viewing in the Q category." k < 0, k can be obtained by fitting a large amount of data. When the user never changes the resolution operation, it is visible. There is no requirement for video clarity for the user; when the user watches a certain type of video, the frequency of increasing the resolution is high, the user is considered to have higher definition of the video, and the initial predicted QoE default user is clear to the video. No requirement, when the user has high expectations for video clarity, the user's experience evaluation will be lower under the same video damage conditions.

M{N = w{N -N . (-l≤M(N _pic )≤l) ( "Γ,"0,,"-1" respectively indicate that the user expresses the emotion of the current video definition as "satisfactory""doesn'tcare""unsatisfied", the absolute value The trend 1 indicates that the higher the emotional level of the user's expression of the current video, the greater the degree of influence on his personal QoE. Among them, N _pic — _m indicates the number of times the resolution is increased when watching the video, N _pic — _de Indicates the number of times the resolution is reduced during the video viewing, which can be obtained by experimental statistics. The brightness of the video currently viewed by the user is judged by the difference between the number of times the user increases the resolution and the resolution is reduced. The initial predicted QoE default user has no requirement for video definition. Under the same video damage condition, the user experience evaluation will naturally As the resolution of the video decreases, it decreases accordingly, and as the clarity of the video increases, it rises accordingly.

Other technical solutions can also accomplish the objectives of the embodiments of the present invention, including but not limited to:

1. The video damage degree can be directly predicted by the network parameters and the application layer parameters, thereby replacing the terminal recording video damage;

2, the initial buffer duration, the re-buffer duration and the level of re-buffering frequency can be more refined; 3. The QoE formula can be fitted according to different users and different video types, and the QoE evaluation of different types of videos by different users can be obtained;

4. The QoE correction method is not limited to linear correction, and the nonlinear exponential function or the logarithmic function can be corrected;

5. The user's behavior analysis can be replaced by direct feedback from the user, but this method will increase the complexity of user operations.

One of ordinary skill in the art will appreciate that all or a portion of the above steps may be accomplished by a program instructing the associated hardware, such as a read-only memory, a magnetic disk, or an optical disk. Alternatively, all or part of the steps of the above embodiments may also be implemented using one or more integrated circuits. Correspondingly, each module/unit in the above embodiment may be implemented in the form of hardware or in the form of a software function module. The invention is not limited to any specific form of combination of hardware and software.

It is a matter of course that the invention may be embodied in various other forms and modifications without departing from the spirit and scope of the invention.

INDUSTRIAL APPLICABILITY The embodiment of the present invention can obtain user information in a timely and effective manner by performing data collection and analysis on the terminal and the server, and has little impact on the user. The embodiment of the present invention fully utilizes the behavior of the user to analyze the user's psychology, including the user. The historical expectation of the video and the emotion when watching the video are compared with the existing methods of obtaining user feedback through the questionnaire survey, and the user's psychology can be obtained without a large amount of manpower and material resources, and is simple and easy to implement.

Claims

claims

1. A mobile streaming media user quality of experience QoE correction method, the method includes: receiving user behavior data and current video impairment data; obtaining the impact value on personal QoE according to the user behavior data; obtaining according to the current video impairment data The initial user quality of experience QoE _mit ; and the modified personal user quality of experience QoE _fmal is obtained according to the impact value on the personal QoE and the initial user quality of experience QoE _mit .

2. The method of claim 1, wherein the impact value on personal QoE includes: a first impact value Et of fluency on the personal QoE and a second impact value Ep of clarity on the personal QoE. ; The step of obtaining the impact value on personal QoE based on the user behavior data includes: establishing a user behavior table based on the user behavior data, and calculating the E _t and E _p based on the user behavior table.

3. The method of claim 2, wherein the user behavior table includes:

L _qtl , L _qtr , L _qfr , respectively represent the average initial buffering duration, average re-buffering duration and average re-buffering frequency when exiting the user's history of viewing (^ type videos;

N _pause , represents the total number of passive pauses of the user watching the video this time; and time, represents the length of the video played; The steps of calculating the E _t and E _p according to the user behavior table include: The E _t =<uID ,

... (C _n ,E _nt )}>; Among them, El _lt ( l<i<n ) represents the impact value of Q-type video fluency on the user's personal QoE, l<E _lt <0;

E _lt +e ₂ M(L _pause ); Among them, I _lt =I(L _qtl ,L _qtr ,L _qfr ) represents the impact of the user's passive exit behavior on QoE; M(L _pause ) represents the impact of the user's passive pause behavior on QoE _The _influence _value _of ₂ M(L _pau

(UlLqti + U Lqfr + Lqtr) I ( _1<1<0 ) . Among them, +U2 +

; M(L _P ause) = e-— , (-1 <M(L _pause )<0) ,

Lizhong _ ^p us ·

^, 'pause, .

time,

The E _p =<uID,{(C ₁ ,E _lp );(C ₂ ,E _2p );...(C _n ,E _np )}>; where, E _ipU≤1≤n) represents category G The impact of video clarity on user’s personal QoE, -1<Ε _ιρ <1;

Eip=lip + etaτΜ N≠c); Among them, _Iip =I( _Fpic — _m ) represents the impact of users’ historical expectations on video clarity on their QoE; M( _Npic ) represents the impact of the video clarity on The impact of user emotions on QoE;

n ₂ respectively represent the user's historical expectations for the clarity of this type of video and the impact value coefficient of current emotions on personal QoE; when the user cold starts, h = 0, at this time Ε _Ψ = niM(N _pic );

H, (-ι ΐφ≤); among them, pic m = ^Ν _Τ - ^ιη N _m pic _c c— _m m— _Ci means that when watching the video of (^ category, the user has performed "add

iV

The number of videos with "large resolution" behavior; Nc, represents the total number of videos in the (^ category that the user watched; k<0;

M(Npic) = W(Npic― in - Npic― de) , (_ J < (N _pic )<l); Among them, N _pic — _m represents the number of times the resolution has been increased while watching the video this time; N _pic — _de represents the number of times the resolution has been reduced while watching the video this time.

4. The method according to claim 1 or 2, wherein the current video impairment data includes: initial buffering duration _Tmi , re-buffering duration! ^^and rebuffering frequency? ^^; The re-buffering duration T _rebuf is the re-buffering duration when the video is automatically paused or automatically exited; if the video buffer can maintain the continued playback of the video, then the pause or exit is determined to be an automatic pause or automatic exit; according to the above The steps of obtaining the initial user quality of experience QoE _mit from the current video impairment data include: obtaining the initial user quality of experience QoE _mit according to receiving the initial buffering duration T _mi , the re-buffering duration T _rebuf and the re-buffering frequency F _rebuf .

5. The method of claim 2, wherein the step of obtaining the modified user quality of experience QoE _fmal based on the impact value on personal QoE and the initial user quality of experience QoE _mit includes: obtaining QoE _{fmal according to the following formula:}

QoE final = QoEinit + JfllEt + TUlEp. Among them, _-l≤Et≤0 , -l<Ep<l, ml + m2 = l, ml and m2 are the weight coefficients of Et _{and Ep} _respectively .

6. A server for mobile streaming user quality of experience QoE correction, the server includes: a data receiving module, which is configured to receive user behavior data and current video damage data; a user behavior recording module, which is configured to receive user behavior data based on the user behavior The data obtains the impact value on personal QoE;

QoE initial evaluation module, which is configured to obtain initial users based on the current video impairment data. Quality of experience QoE _mit ; and

The QoE correction module is configured to obtain the corrected user quality of experience QoE _fmal based on the impact value on personal QoE and the initial user quality of experience QoE _mit .

7. The server according to claim 6, wherein the impact value on personal QoE includes: the first impact value of fluency on the personal QoE

The second impact value Ep of Et and clarity on the personal QoE; the user behavior recording module is configured to obtain the impact value on the personal QoE based on the user behavior data in the following manner: Based on the user behavior data, establish User behavior table, the E _t and _Ep are calculated according to the user behavior table.

8. The server according to claim 7, wherein the user behavior table includes:

L _qtl , L _qtr , L _qfr , respectively represent the average initial buffering duration, average re-buffering duration and average re-buffering frequency when exiting the user's historical viewing (^ type videos; N _pause , represents this viewing The total number of passive pauses of the video user; time, represents the length of the video played; The user behavior recording module is set to calculate the E _t and E _p according to the user behavior table in the following manner: The E _t =< uID,

... (C _n ,E _nt )}>; Among them, E _ltu≤1≤n) represents the impact value of Q-type video fluency on the user's personal QoE, l<E _lt <0;

E _it = I _it + _e2 M (L _p where, I _lt =I(L _qtl ,L _qtr ,L _qfr ) represents the impact of the user's passive exit behavior on QoE; M(L _pause ) represents the impact of the user's passive pause behavior on QoE influence value; _{ei + e2} = i, ei, e ₂ respectively represent the user’s passive The influence value coefficient of exit behavior and passive pause behavior on the QoE of this video viewing; When the user cold starts, ⁰ , at this time E _it = _e2 M(L _pause );

I _it = -1 + (inLqti + uiLqfr + mLqtr) 13 ( _i<i. _t <0 ) . Among them, «l +«2+«3 <1;

M ( e-— - 1 , (-1 <M(L _pause )<0) , T = ^ ^pause in _JL ·

time the E _p =<uID, {((^,E _1ρ ); (0 ₂ ,E _2ρ ); ... (C _n ,E _np )}>; where, E _{ip (1≤1≤n} , Indicates the impact value of a certain type of video Q definition on the user's personal QoE, -1<Ε _ιρ <1;

Among them, I _ip =I(F _pic — _m ) represents the impact of users' historical expectations on video clarity on QoE; M(N _pic ) represents the impact of user emotions caused by video clarity on QoE; 1^ +112=1,1^, n ₂ respectively represent the user's historical expectations for the clarity of this type of video and the impact value coefficient of current emotions on personal QoE; when the user cold starts, p = 0, at this time Ε _Ψ = H2M( Np,c);

H ( _-i≤iiP≤o );

where ₌ N _{vlc m Ci}

¹ pic in Ν, pic_m_c; ^r represents the number of videos in which the user has performed the "increase resolution" behavior when watching videos in the (^ category; N. represents the total number of videos in the (^ category that the user has watched; k<0;

M(Npw) - w N _P ic _ in - Npic _ de) (_ j <Μ(Ν■ )< 1 ) · Among them, Ν _ρκ : — _ιη represents the number of times the resolution is increased when watching the video this time; N _pic — _de represents the number of times the resolution was reduced when watching the video this time.

9. The server of claim 6, wherein, The current video damage data includes: initial buffering duration T _mi , re-buffering duration T _rebuf and re-buffering frequency F _rebuf ; the re-buffering duration T _rebuf is the re-buffering duration when the video is automatically paused or automatically exited; if If the video cache area can maintain the continued playback of the video, the pause or exit is judged to be automatic pause or automatic exit; the QoE initial evaluation module is set to obtain the initial user experience quality QoE _mit based on the current video damage data in the following manner: Receive all The initial buffering duration T _mi , the re-buffering duration T _rebuf and the re-buffering frequency F _rebuf are used to obtain the initial user experience quality QoE _imt .

10. The server according to claim 7, wherein the QoE correction module is configured to obtain the corrected user quality of experience QoE _final according to the impact value on personal QoE and the initial user quality of experience QoE _imt in the following manner: QoE _fmal is obtained according to the following formula:

QoEfinal = QoEinit + JfllEt + JfllEp. Among them, _-l≤Et≤0 , -1<Ερ<1, ml + m2 = \, ml and m2 are the weight coefficients of Et _and _Ep respectively.

11. A computer program, including program instructions, which when executed by a server, enable the server to execute the method described in any one of claims 1-5.

12. A carrier carrying the computer program of claim 11.