Fadya Abbas, Nahrain University, Iraq

Dealing with extensive amounts of textual data requires an efficient deep learning model to be adapted. However, the following reasons; the highly ambiguous and complex nature of many prosodic phrasing also enough dataset suitable for system training is always limited, cause big challenges for training the NLP models. This proposed conceptual framework aims to provide an understanding and familiarity with the elements of modern deep learning networks for NLP use. In this design, the encoder uses Bidirectional Long Short-Term Memory deep network layers, to encode the test sequences into more context-sensitive representations. Moreover, the attention mechanism is mainly used to generate a context vector that is determined from distinct alignment scores at different word positions, hence, it can focus more on a small words' subset. Hence, the attention mechanism improved the model data efficiency, and the model performance is validated using an example of data sets that show promise for a real-life application.

NLP, Deep Learning, LSTM, Attention Mechanism, Data Efficiency.