Building conversational Question Answer Machine and comparison of BERT and its different variants

Older search engines and question answering machines could only return results from similarly themed websites. Modern question answering machines and systems may provide in-depth solutions to all questions, thanks to breakthroughs in fields like machine reading comprehension, transfer learning, and language modelling. In this paper, I shall add in Machine Reading Comprehension and describe its operation. Machine reading comprehension is the problem of creating a system that can understand text and respond appropriately to questions about that content. When used for reading comprehension, the model takes as inputs a question and a context or excerpt. The model's output is the conclusion reached after analysing the data. Existing Question Answering systems answer to each Question independently, which increases repetition because the entity is repeated regardless of whether the current Question is a follow-up one or not. The BERT like transformer deep learning model that I'll be employing is based on the theories of transformers; it was developed by Google researchers in 2018; and it employs dynamic weightings depending on the links between output and input parts. SQuAD 1.1 and SQuAD 2.0 are just two examples of the high-performance datasets now on the market. I am free to use any of these datasets to train my model, however the publicly available Oracle technical documentations will act as my validation set. As this is the case, my model will study the docs to answer user questions. Customers can use this question answering machine instead of resorting to a Google search to receive the information they need.