Table 1.
Sample of challenging sequences.
Fig 1.
Complete process of dataset preparation.
Table 2.
Article statistics.
Table 3.
Sentence structure conversion by separating punctuation.
Table 4.
Sentence after tokenization and set up sentence indicator.
Table 5.
Tagging scheme.
Table 6.
Sample of NER dataset.
Fig 2.
Named entity vs. non-named entity ratio.
Fig 3.
Ratio of named entities.
Fig 4.
Frequency of named entities.
Table 7.
Dataset statistics.
Table 8.
Train and test data distribution.
Fig 5.
Workflow diagram.
Fig 6.
Type-1 model architecture.
Fig 7.
Type-2 model architecture.
Table 9.
Statistics of matched words in each word embedding.
Table 10.
Changing nature of the root words with the addition of suffixes and prefixes.
Fig 8.
Character level feature extraction with CNN.
Table 11.
Example sequence for prediction.
Table 12.
Sample evaluation.
Table 13.
Exact and partial precision scores.
Table 14.
Exact and partial recall scores.
Table 15.
Exact and partial F1 scores.
Table 16.
Micro F1 scores.
Table 17.
Comparison of hybrid and non-hybrid models’ F1 scores.
Table 18.
Comparison of results between four word embeddings.
Table 19.
Comparison of results on integrating CNN layer.
Table 20.
Comparison of results on integrating CRF layer.
Table 21.
Comparison of results between BGRU and BLSTM based models.
Fig 9.
Dropout effect.
Table 22.
Count of correctly predicted NE classes.
Table 23.
Test sample-1.
Table 24.
Test sample-2.
Fig 10.
Comparison of results between four models.
Table 25.
Macro and micro F1 scores of the best model.
Fig 11.
The graphical representation of confusion matrix of the best performing model.
Table 26.
Results on different challenges for BGRU+CNN+CRF (word2vec(cbow)).
Table 27.
Results of other existing model along with our proposed model for Bengali NER.
Table 28.
Dataset details of other existing models along with our proposed model for Bengali NER.