Optimization of Natural Language Understanding with Contextual Embeddings
R. Sangeetha, Durgesh Srivastava, J. Logeshwaran, Pramod Vishwakarma, Satvik Vats
Abstract
Natural Language Understanding (NLU) has recently made considerable progress, but there is still an immediate need to improve its performance. To this end, researchers have addressed the issue by introducing contextual embedding’s, which enable the NLU model to map words to their contextual meanings rather than just looking at their individual meanings. Contextual embedding’s enable the model to capture the nuances of words in the various contexts they are used in, allowing for better understanding and performance. Two methods— feature engineering and transfer learning—have been employed to further improve performance. With feature engineering, transformed features are used to obtain improved accuracy and faster training times whereas transfer learning uses pre-trained models to reduce the computational power required for training. This approach has resulted in improved accuracy in the various language understanding tasks. Furthermore, the innovative use of contextual embedding’s in combination with various optimization methods has resulted in a much more reliable and accurate NLU model.