Word segmentation class using the maximum probability approach. Still in verifying now.
PS, I found that when using the same lexicon, the difference between the result segmented by maximum probability and reverse maximum match approach is less than 1%.