Thai Name Gender Classification using Deep Learning
Chattarin Horhirunkul, Sangsuree Vasupongayya, Suthon Sae-Wong, Seksun Suwanmanee, Touchai Angchuan · 2021
Thai name gender prediction model is presented in this work. The proposed model is based on a Bidirectional Long Short Term Memory (Bi-LSTM). Two dropout blocks are added to reduce the overfitting issue of the model. Two sets of Thai name corpus are used for training and evaluating the proposed model. The first corpus consists of Thai names collected from social media in this work while the second corpus is a combination of the first corpus and the publicly available dataset. The names are classified into male and female. The unisex classification is labeled to the names that the proposed model cannot decide to be either male or female with a probability greater than 0.65. The proposed model provides an average overall accuracy of78.65%. The model generated from the second corpus provided the best performance at an average overall accuracy of 81.43%. The model generated from the first corpus provided the average overall performance of 75.87%.