【摘要】：基于深度學(xué)習(xí)的循環(huán)神經(jīng)網(wǎng)絡(luò)方法,面向中文字和詞的特點(diǎn),重新定義了地名標(biāo)注的輸入和輸出,提出了漢字級(jí)別的循環(huán)網(wǎng)絡(luò)標(biāo)注模型.以詞級(jí)別的循環(huán)神經(jīng)網(wǎng)絡(luò)方法為基準(zhǔn),本文提出的字級(jí)別模型在中文地名識(shí)別的準(zhǔn)確率、召回率和F值均有明顯提高,其中F值提高了2.88%.在包含罕見(jiàn)詞時(shí)提高更為明顯,F值提高了26.41%.
[Abstract]:Based on the circular neural network method of deep learning and the characteristics of Chinese characters and words, the input and output of place names are redefined, and a Chinese character level circular network annotation model is proposed. Based on the word-level cyclic neural network method, the accuracy, recall rate and F value of the word-level model in Chinese geographical names recognition are obviously improved, in which the F value is increased by 2.88. The value of F was increased by 26.41 when the rare words were included.
【作者單位】： 南京理工大學(xué)經(jīng)濟(jì)管理學(xué)院;計(jì)算機(jī)軟件新技術(shù)國(guó)家重點(diǎn)實(shí)驗(yàn)室(南京大學(xué));
【基金】：國(guó)家自然科學(xué)基金資助項(xiàng)目(71503124,71303120) 江蘇省社會(huì)科學(xué)基金資助項(xiàng)目(15TQC03)
【分類(lèi)號(hào)】：TP18;TP391.1
，

本文編號(hào)：2307210