[Paper Review] Predicting gender of Brazilian names using deep learning.

Rosana C. B. Rego, Verônica Maria Lima Silva|arXiv (Cornell University)|Jun 18, 2021

Authorship Attribution and Profiling26 references4 citations

TL;DR

This paper evaluates feedforward and recurrent deep neural networks—MLP, RNN, GRU, CNN, and BiLSTM—for predicting the gender of Brazilian first names. Using a Brazilian name dataset, it demonstrates that recurrent models, particularly BiLSTM, outperform feedforward models, achieving over 90% accuracy by treating names as sequential string features.

ABSTRACT

Predicting gender by the name is not a simple task. In many applications, especially in the natural language processing (NLP) field, this task may be necessary, mainly when considering foreign names. Some machine learning algorithms can satisfactorily perform the prediction. In this paper, we examined and implemented feedforward and recurrent deep neural network models, such as MLP, RNN, GRU, CNN, and BiLSTM, to classify gender through the first name. A dataset of Brazilian names is used to train and evaluate the models. We analyzed the accuracy, recall, precision, and confusion matrix to measure the models' performances. The results indicate that the gender prediction can be performed from the feature extraction strategy looking at the names as a set of strings. Some models accurately predict the gender in more than 90% of the cases. The recurrent models overcome the feedforward models in this binary classification problem.

Motivation & Objective

• To investigate the effectiveness of various deep learning architectures in predicting gender from Brazilian first names.
• To evaluate whether recurrent models like GRU and BiLSTM outperform feedforward models such as MLP and CNN in this binary classification task.
• To analyze model performance using standard metrics including accuracy, precision, recall, and confusion matrices.
• To explore the feasibility of gender prediction using only the name as input, treating it as a sequence of characters.
• To determine the optimal deep learning architecture for gender classification on Brazilian names based on empirical evaluation.

Proposed method

• Employed a dataset of Brazilian first names labeled with gender for training and evaluation.
• Implemented feedforward models (MLP, CNN) and recurrent models (RNN, GRU, BiLSTM) for gender classification.
• Treated each name as a sequence of characters, using word embedding or one-hot encoding for input representation.
• Trained models using standard deep learning optimization techniques with categorical cross-entropy loss.
• Evaluated models using accuracy, precision, recall, and confusion matrices to compare performance.
• Applied feature extraction strategies that focus on character-level patterns in names to capture linguistic cues for gender prediction.

Experimental results

Research questions

• RQ1 Which deep learning architecture performs best for gender prediction from Brazilian first names?
• RQ2 How do recurrent models compare to feedforward models in terms of accuracy and robustness for this task?
• RQ3 To what extent can gender be predicted using only the name as input through sequence-based feature learning?
• RQ4 What are the precision and recall trade-offs across different models in this binary classification problem?
• RQ5 Can character-level sequence modeling effectively capture gender-associated patterns in Brazilian names?

Key findings

• Recurrent models, particularly BiLSTM, outperform feedforward models in gender prediction accuracy on Brazilian names.
• The best-performing models achieve a gender prediction accuracy of over 90% on the dataset.
• GRU and BiLSTM models show higher recall and precision compared to MLP and CNN, indicating better generalization.
• The confusion matrix analysis reveals that recurrent models make fewer misclassifications, especially for names with ambiguous gender cues.
• Feature extraction based on character-level sequences effectively captures linguistic patterns associated with gender in names.
• The study confirms that deep learning models can reliably predict gender from names using only the name string as input.