Named entity identification method based on time convolution network

Abstract

The invention relates to a named entity identification method based on a time convolution network. The method comprises the following steps of: firstly, constructing a feature representation layer which mainly consists of a word vector and a character feature layer, wherein the word vector layer and the character vector layer respectively accept words and characters as input, and respectively mapdiscrete One-hot representations to respective continuous dense low-dimensional feature spaces; splicing the word vectors and the character-level vectors to represent features of the words in a particular semantic space; secondly, taking the spliced features as input of a time convolution network, extracting different features through the time convolution network with different fusion convolutionkernel sizes, and obtaining final features h1h2... hn; finally, taking the obtained features as input of a CRF layer; and after the CRF further restrains context annotation, outputting sequence annotation results y1y2... yn. Compared with an existing LSTM network, the TCN network has the advantages that the recognition precision is slightly improved, and the training time is only about 1/3 of thatof the LSTM network.

Images