Named entity recognition refers to the recognition of the specified type of entity in the text to be processed.In general,the types of entities that need to be identified for named entity recognition are divided into three categories: entity,number,and time,or further subdivided into seven categories: person name,institution name,place name,time,date,currency,and percentage.And in a particular domain,the various entity types within the domain are defined accordingly.Named entity recognition is a fundamental key task in natural language processing.It paves the way for many natural language processing tasks such as relationship extraction,event extraction,knowledge maps,machine translation,and question answering systems.It's important in the process of natural language processing technology becoming practical.Character-level Chinese named entity recognition using Long Short-Term Memory(LSTM)to merge dictionaries has achieved great success.However,in terms of computational efficiency,the recurrent neural network(RNN)processes sentences in a sequential manner and cannot take full advantage of GPU parallelism.In contrast,the Convolutional Neural Network(CNN)can take full advantage of GPU parallelism through its feedforward architecture.This paper proposes a Chinese named entity recognition method Hownet+Simplified the Usage of Lexicon+Rethinking CNN(HSLR-CNN): a CNN-based method that resolves potential word conflicts through rethinking mechanism,preserves all the possible words through incorporating the lexicon information into the vector representations of characters to reduce word segmentation errors,reduces Chinese polysemy by adding external language knowledge How Net,and uses GPU parallelism and chained input to improve computing efficiency.Experimental results on the Resume dataset and Weibo dataset show that the method can achieve better performance than the baseline method,and can saving training and testing time costs greatly. |