Based On Maximum Entropy Model Of Chinese Named Entity Recognition

This dissertation mainly researchs means of the name entity recognize via designing and implementing a system which consider name,placename and organization as core.The major work and feature of this dissertation are as follows: 1 , Regarding maximum entropy as the basic frame , based on the marked corpus , setting up the language model of the maximum entropy and not using artificial dictionary . Utilize the part decodes algorithms and viterbi algorithms to recognize name and placename in word level and orgnizaton in words level. 2, We present a tree-trellis decode arithmetic for the decode of maximum entropy.The advantage of this algorithm lies in it can get the best result and the N best result with the linear time complexity that increases by the length of the text; Can judge whether the adjoint state is legal , has solved the potential conflict problem in marking the result.3 , In order to improve the performance of the model,We have tried different feature template,and show the contrast data.The result of experiment indicates that the name entity recognize based on maximum entropy in the condition of relatively little corpus can achieve good performance.