Machine Reading Comprehension

This chapter provides an overview of machine reading comprehension (MRC), a natural language processing task where a computer system analyzes an article to answer questions about it. The document focuses on implementing a Chinese MRC system using the ERNIE 3.0 pre-training model and the extractive Q&A method, which involves identifying the exact start and end positions of an answer within a given text. It details the entire process, including data processing using the DuReaderrobust dataset, model construction with a linear layer to predict answer positions, training configuration with the AdamW optimizer and cross-entropy loss, and model evaluation using F1 score and overall exact match as metrics. The chapter concludes with a demonstration of how the trained model can predict answers for new input texts.