| Summary: | Neural Network models are considered universal learning due to their powerful ability to solve text classification with different data types. Hybrid neural networks in text classification using bidirectional Long Short-term Memory (BiLSTM), Convolutional Neural Network (CNN) and bidirectional Gated Recurrent Unit (BiGRU) have been widely used. In this study, we conducted a preliminary study on text classification of social bias data by combining the CNN, Bidirectional LSTM, Bidirectional GRU, and attention mechanism. The output from bidirectional architectures are given a distinct focus through attention mechanisms employed in both layers. Social bias is opinionated data which is often demonstrated in stereotypical and prejudiced behavior, presenting unfair decision including hatred and emotion about people's identity such as race, religion, ethnicity and gender. Social bias is considered under low resource scenario whereby the data size is limited and subjective. The result shows the proposed hybrid model gives promising results and good potential in learning a small size of data compared to single models. © 2023 IEEE.
|
|---|