Skip to main content

Table 3 Comparison between a series of CLSTM (contextual long short-term memory networks [LSTMs] with conditional random fields [CRF]) experiments and the comparative methods on the NCBI, GM, and CDR corpora using strict matching

From: Biomedical named entity recognition using deep neural networks with contextual information

Strict matchingNCBIGMCDR
ModelTrial #prfprfprf
BiLSTM-78.9182.6080.7172.2272.4472.3383.5680.2681.88
BiLSTM-CRF-82.1984.5883.3780.7979.8180.3087.5283.5885.50
GRAM-CNN-84.4583.9284.1880.2378.8379.5386.0885.4985.79
BERT-81.0780.7380.9081.7281.5981.6586.2185.2385.72
CLSTM (word+char levels)184.7386.6785.6881.7581.1481.4487.2585.6686.44
 284.4385.8385.1281.2680.6780.9687.1685.4086.27
 386.1884.4885.3282.0780.2481.1487.9384.5686.21
 485.5685.2185.3982.9779.6681.2887.7185.1786.42
 584.6285.4285.0281.0280.7080.8688.2784.3686.27
CLSTM average 85.1085.5285.3181.8180.4881.1487.6685.0386.33