Data-to-text system to summarise time series data
Development of a data-to-text system to summarise time series data chronologically. The summary describes the time series based on segments and their associated characteristics, namely trend direction, duration and variability. This is achieved by using change point detection algorithms for segmenting the time series and fuzzy set theory to extract linguistic variables of the segments' characteristics. The generalisation ability and speed of the system is tested with stock exchange data. The change point detection algorithms precision if data is very volatile. In terms of speed, approximate change point detection models outperform optimal methods greatly and compute results within seconds.