The corpus analysis methods in Chinese keyword extraction look on the corpus as a single sample of language stochastic process. But the distributions of keywords in the whole corpus and in each document are very different from each other. The extraction based on global statistical information only can get significant keywords in the whole corpus. Max-duplicated strings contain the local significant keywords in each document. In this paper, we designed an efficient algorithm to extract the max-duplicated strings by building PAT-tree for the document, so that the keywords can be picked out from the max-duplicated strings by their SIG values in the corpus.
展开▼
