Generating Linguistic Expression of Charts Based on Comparison of Multiple Time-Series Data

Abstract

This paper proposes a method for generating linguistic expressions from a time- series data. The proposed method takes differences and similarities among multiple time-series data into consideration: The method generates linguistic expressions by executes three processes sequentially. First, a characteristic such as "rise," "drop," and "stable" is evaluated in each data point of the data series. Second, for each data point in a data series, a weight is assigned by calculating a degree of attention, which is estimated by comparison with another time-series data. Finally, the most pertinent expression is selected.