摘要 |
<P>PROBLEM TO BE SOLVED: To suppress the increase of computational complexity or memory usage when detecting a trend from a data stream. <P>SOLUTION: The trend detection device calculates the similarity scores of the partial sequence of a data stream and a comparison pattern segmented from the data stream, and records the similarity scores and the start position (t<SB>s</SB>, i<SB>s</SB>) of the partial sequence in a time warping matrix. When the similarity scores are equal to or more than a predetermined threshold ε, and the partial sequence with predetermined sequence length is detected, this is detected as a trend. Also, when the similarity scores are lower than the predetermined threshold ε, and the partial sequence with the predetermined sequence length is detected, this is defined as a new comparison pattern, and a trend is detected by using the new comparison pattern afterwards. <P>COPYRIGHT: (C)2009,JPO&INPIT |