Negative sequential patterns (NSP) are critical and sometimes much more
informative than positive sequential patterns (PSP) in many intelligent
systems and applications. However, the existing NSP algorithms do not allow
negative items being contained in an element except the NegI-NSP algorithm,
which can obtain many meaningful sequences with negative items in an
element. NegI-NSP, however, hasn?t considered the following problems: (1) it
uses a single minimum support to all size sequences, which is unfair to a
long size sequence; (2) it only mines NSP from PSP, not from infrequent
positive sequences (IPS), which also contain many useful NSP. So we propose
an efficient algorithm, named MLMS-NSP, to mine NSP based on multiple level
minimum supports (MLMS) from PSP and IPS. Firstly, MLMS scheme is proposed
by assigning different minimum supports to sequences with different sizes.
Secondly, IPS are constrained by combining MLMS, and then the NSP is
obtained from these IPS. Finally, experimental results show that the
MLMS-NSP algorithm can effectively mine NSP from IPS, and the time efficiency
is higher than using single minimum support.