Named Entity Recognition For Chinese Microblog With Deep Learning

Following the rise of Personal Media like Weibo which kind of micro-blog,the public enthusiasm for the media industry is also growing.There are more and more researchers study on Weibo data and configure how to use its' worthy data.The traditional text of natural language processing development has a very mature stage,but in the micro-blog text of natural language processing is just in its infancy,and the traditional machine learning method has proven to can't complete copy.This article is written in Chinese micro-blog research,namely the named entity recognition which is a part of the natural language processing research.In order to complete the named entity recognition on micro-blog data,we first use of traditional machine learning methods such as Hidden Markov and Conditional Random Field,however,the experimental effect is poorer.We summarized the reason why traditional machine learning has such a poor performance in micro blog,and the deep learning method was proposed to accomplish named entity recognition,it mainly includes the following aspects:1,Propose a normalized process to the micro-blog text.Considering on the difference between micro-blog text with the traditional text,we will build a system to normalize the micro-blog text,and prepare for the next step to classify our named entity.2,Propose a novel Convolutional Context Window Network for Named Entity Recognition on micro-blog text.Considering the limitations that using the traditional machine learning to do the micro-blog text's Named Entity Recognition work,and the huge difficulty to select the feature by hand,we proposed a model that based on deep learning to study the named entity recognition work.First of all,we apply a convolutional network to a micro-blog text which the model was mainly used for image classified before.Then we improve the structure by changing the activation function of CNN model and using Dropout to prevent overfitting.Compared with traditional machine learning methods,our Convolution Context Window Network(CCWN)has better performance,3,Propose a hybrid model for Named Entity Recognition on micro-blog text.Considering the complexity of the convolutional neural network training,We introduce a Hybrid model(GRU-CNNs)that not only include our CCWN,but also use a Variants of Recurrent Neural Network named Gated Recurrent Unit(GRU).The GRU-CNNs model got a better recognition performance.