Multi-dimensional sparse time series: feature extraction

Multimedia sequential data represent the behavior of multiple measurements on some process and may be analyzed as multi-dimensional time series via entropy and statistical linguistic techniques. We introduce three markers: influence area, consistency and diversification. The former two refer to the quality of the dynamic change of the data with time; the last one measures the variability of recurrent patterns. These markers are useful in classification or clustering of large databases, prediction of future behavior and attribution of new data. We show an application concerning different investment strategies in purchasing commercials in advertising market.