Extraction of Frequent Sequential Patterns from Sequence at Uneven Intervals
Kenta Morita, Haruhiko Takase, Masachika Sawamura, Naoki Morita, Hidehiko Kita · 2018
In this paper, we discuss a method to extract frequent sub-sequences from time series data with variations in intervals of input elements. Several methods for extracting frequent sub-sequences from sequence data have been proposed. However, since these existing methods target series data whose element input intervals are constant, it is difficult to extract from time series data whose input intervals are not uniform, such as voice data. Therefore, we propose a method that can extract frequent sub-sequences from time series data in which elements and element appearance intervals are not uniform. To extract frequent sub-sequences from time series data whose element appearance intervals are not uniform, we proposed a method to utilize the internal potential characteristic of the unit of the spiking neural network. To confirm the effectiveness of the proposed method, we created time series data with nonuniform intervals of elements and elements similar to Japanese voice data and fed it to the proposed neural network. As a result, the proposed neural network extracted frequent sub-sequences from time series data with the same performance as extraction from sequence data.