Cognitive Computing with Deep Neural Networks

This study introduces a novel approach to enhancing question answering (QA) systems, employing a RoBERTa based architecture and complexity-enhanced input features. The research is structured into four main components: training methodology, feature engineering, model construction, and data preprocessing. A Python function is proposed to calculate linguistic difficulty metrics for input sentences using readability measures and natural language processing techniques. Tensor-Flow Datasets (TFDS) are then utilized to load and preprocess the SquAD (Stanford Question Answering Dataset) dataset, facilitating efficient training. To provide contextual information to the input representation, word embeddings from pre-trained GloVe vectors are integrated with complexity metrics to prepare input features. Incorporating these enriched features, a distinctive question-answering model based on the RoBERTa architecture is trained using the AdamW optimizer and Cross Entropy Loss. The training process employs iterative epochs to optimize the model’s parameters and minimize the loss function. An independent validation dataset is employed to assess the model’s performance, demonstrating the effectiveness of the proposed method in enhancing the accuracy and robustness of the QA system. In conclusion, this study presents a systematic approach to enhancing system quality by combining state-of-the-art neural architecture with input properties augmented by complexity.