The hot phrases in the social network text streams can reflect the hidden hot topics and sudden events.This paper proposes a hot phrase mining technology which can support various hot degree measures without word segmentation.We first construct an AC-Trie using the candidate phrases gathered from text streams.Based on such AC-Trie

we record the historical occurrence frequency of phrases on the Trie by scanning the following streams in single-pass.Furthermore

the AC-Trie needs to be reconstructed using the new samples in the text stream because of the evolution of hot phrases.Thus

we start the reconstruction dynamically according to estimating the occurrence frequency of the missed phrases.The experiments on the Sina micro-blog show that our approach is effective (precision of 89%) and efficient (overhead is 2% of nave approach).