| Chinese word segmentation is one of the basic works of Chinese information processing.Nowadays,although traditional Chinese word segmentation methods have become matured,but these traditional methods still have some limitations in the automatic construction of feature engineering and domain adaption.Manual feature engineering is time-consuming and costly,and it is difficult to ensure that the extracted features can be used to cover each language phenomena.In recently years,with the development of deep neural network(DNN)technologies in natural language processing tasks,these methods,which do not need to manually formulate rules and extract feature templates,it is becoming more and more rapidly favored by researchers.On the one hand,these DNN based methods can automatically extract features and autonomously learn the internal language rules from the annotation corpus.On the other hand,the precision of Chinese word segmentation is affected by a large number of the out-of-vocabulary(OOV)words,and most of the OOV words are contented by named entities and terminologies.Therefore,this paper adopted a DNN based method,which uses Fixed-size Ordinally Forgetting Encoding(FOFE)to integrate the context information of the sequence into the word segmentation model,and this method can be used to develop an universal sequence labeling system for word segmentation,named entity recognition and terminology extraction.The effectiveness of our proposed method was experimentally verified according to our experiments based on various annotation datasets.The innovations points and main research achievements of the paper are shown as follows:(1)Design a framework of using FOFE to integrate context information into the framework of the sequence tagging method,and it can be used to develop an universal method of data preprocessing,model training and testing;(2)Use a variety kinds of different Chinese word segmentation annotation datasets to prove the effectiveness of word segmentation and named entity recognition;(3)Analyze the patent literature and its technical terms,devise and verify the patent text labeling scheme using by our proposed framework.The effectiveness of our proposed methods has been proved according to the experiments of patent terminology extraction. |
