TY - GEN
T1 - Trend analysis and prediction in multimedia-on-demand systems
AU - Ng, Danny M. P.
AU - Wong, Eric W. M.
AU - Ko, K. T.
AU - Tang, K. S.
PY - 2001/6
Y1 - 2001/6
N2 - Resource-demanding services such as Multimedia-on-Demand (MOD) become possible as broadband Internet is getting more popular. However, as the size of multimedia files grows rapidly, storage of such large files becomes a problem. Since multimedia contents will generally become less popular with time, it is desirable to design a prediction algorithm so that the multimedia content can be unloaded from the server when it becomes unpopular. In this paper, we have two objectives: 1) analyse the MOD viewing trend in order to understand the viewing behaviour of users, 2) predict the viewing trend based on the knowledge obtained from the trend analysis. For trend analysis, we study three traditional regression models, including linear regression, exponential regression, and power regression, and propose two additive regression models, exponential-exponential-sum (EES) and exponential-power-sum (EPS), to improve the goodness of fit. Then, the most fitted models will be used in trend prediction. Four prediction app roaches, Fixed Regression Selecting (FRS), Continuous Regression Updating (CRU), Historical Updating (HU), and Continuous Regression with Historical Updating (CRHU) are proposed. From the numerical results, we find that CRHU, which is constructed by considering historical trend and new incoming viewing request data, is in general the best method in forecasting MOD trend. © 2001 IEEE
AB - Resource-demanding services such as Multimedia-on-Demand (MOD) become possible as broadband Internet is getting more popular. However, as the size of multimedia files grows rapidly, storage of such large files becomes a problem. Since multimedia contents will generally become less popular with time, it is desirable to design a prediction algorithm so that the multimedia content can be unloaded from the server when it becomes unpopular. In this paper, we have two objectives: 1) analyse the MOD viewing trend in order to understand the viewing behaviour of users, 2) predict the viewing trend based on the knowledge obtained from the trend analysis. For trend analysis, we study three traditional regression models, including linear regression, exponential regression, and power regression, and propose two additive regression models, exponential-exponential-sum (EES) and exponential-power-sum (EPS), to improve the goodness of fit. Then, the most fitted models will be used in trend prediction. Four prediction app roaches, Fixed Regression Selecting (FRS), Continuous Regression Updating (CRU), Historical Updating (HU), and Continuous Regression with Historical Updating (CRHU) are proposed. From the numerical results, we find that CRHU, which is constructed by considering historical trend and new incoming viewing request data, is in general the best method in forecasting MOD trend. © 2001 IEEE
KW - Additive regression
KW - Continuous regression updating
KW - Continuous regression with historical updating
KW - Fixed regression selecting
KW - Historical updating
KW - Multimedia-on-demand
KW - Regression models
KW - Trend analysis
KW - Trend prediction
UR - http://www.scopus.com/inward/record.url?scp=0034873502&partnerID=8YFLogxK
UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-0034873502&origin=recordpage
U2 - 10.1109/ICC.2001.936905
DO - 10.1109/ICC.2001.936905
M3 - RGC 32 - Refereed conference paper (with host publication)
SN - 0-7803-7097-1
VL - 4
T3 - IEEE International Conference on Communications
SP - 1292
EP - 1298
BT - ICC2001 The IEEE International Conference on Communications
PB - IEEE
T2 - International Conference on Communications (ICC2001)
Y2 - 11 June 2001 through 14 June 2001
ER -