A multi-Cross Matching Network for Chinese Named Entity Linking in Short Text

Abstract

Named entity linking(NEL) for Chinese short text is regard as a ranking task. Current research on NEL is mostly about long text and rich language, which may unfit Chinese short text situation. We propose a multi-cross matching network(CMN) to solve these problems. CMN first matches a candidate and the mention with the text in three different degree, and extracts important matching information through a couple of convolution and pooling layers. Then sequence vectors are accumulated through a Gated Recurrent Unit(GRU) which distill relationships among words in texts. Finally we use dynamic average algorithm to calculate the mention-entity similarity. Experimental results demonstrate that CMN exhibits the outstanding performance on several Chinese datasets for the Chinese short text entity linking problem.