CharTeC-Net: An Efficient and Lightweight Character-Based Convolutional Network for Text Classification

Table 2

Accuracy results for all the models. Numbers are in percentage.

Models	AG’s news	Sogou news	DBPedia	Yelp review polarity	Yelp review full	Yahoo! answers	Amazon review full	Amazon review polarity
Bag of words [19]	88.8	92.8	96.6	92.2	58.0	68.9	54.6	90.4
ngrams [19]	92.0	97.0	98.6	95.6	56.3	68.5	54.3	92.0
ngrams TFIDF [19]	92.4	97.2	98.7	95.4	54.8	68.5	52.4	91.5
Char-CNN [19]	87.2	95.1	98.3	94.7	62.0	71.2	58.7	94.5
fastText [8]	92.5	96.8	98.6	95.7	63.9	72.3	60.2	94.6
Char-CRNN [24]	91.4	95.2	98.6	94.5	61.8	71.7	59.2	94.1
VDCNN [22]	91.3	96.8	98.7	95.7	64.7	73.4	63.0	95.7
Naive Bayes [30]	90.0	86.3	96.0	86.0	51.4	68.7	—	—
Kneser-Ney Bayes [30]	89.3	94.6	95.4	81.8	41.7	69.3	—	—
MLP Naive Bayes [30]	89.9	76.1	87.2	73.6	40.4	60.6	—	—
Discriminative LSTM [30]	92.1	94.9	98.7	92.6	59.6	73.7	—	—
Generative LSTM-independent comp. [30]	90.7	93.5	94.8	90.0	51.9	70.5	—	—
Generative LSTM-shared comp. [30]	90.6	90.3	95.4	88.2	52.7	69.3	—	—
Char shallow-and-wide CNN [23]	90.7	—	98.0	94.4	60.3	70.2	—	—
Word shallow-and-wide CNN [23]	92.2	—	98.7	95.9	64.9	73.0	—	—
CharTeC-Net (our model)	91.6	96.4	98.5	95.83	63.6	67.8	61.5	94.0