Research Of Named Entity Recognition Based On Probabilistic Dependency

With more and more text information are easier to access,it is increasingly important to get simple and effective information from this complex data in the era of bag data explosion.Named Entity Recognition is a problem of locating and classifying import proper nouns in text,such as person name,location name.In Chinese,the named entity is the significant part of the sentence,it is useful to find and extract the named entity.Text data is independent to each other.This paper uses data as a set of random variables,and then to mine the probability dependence of entities and other information in random variables.Considering the particularity of the named entity recognition task,we start from the dependency parsing and subsequences,and propose an improvement algorithm on named entity recognition.The main work of this paper is as follows.(1)The neural network structure of fusion dependent syntax information is proposed to identify the named entity,and the effect of the two-way information transmission can be obtained through the transmission of one-way information.This method takes into account the semantic relationship between the information related to the physical position in the sentence obtained by the syntactic analysis as a feature to improve the recognition performance of the named entity.This paper proposes to convert the word level dependency-based syntax analysis into character-based dependency syntax information.To get more extra information and reduce the impact of incorrect dependency parsing results.The tree long short term memory neural network is introduced to mine the tree-like information of the dependent syntax analysis.(2)A named entity recognition method based on neural network semi-Markov conditional random field is proposed.This method solves the local dependency problem caused by conditional random field,which leads to the improvement of final recognition performance.