Open-domain question answering
Question-answering (QA) systems aim to emulate the human process of searching for information online, with machine learning methods employed to improve the accuracy of the provided answers. In this recipe, we will demonstrate how to use RNNs to predict long and short responses to questions about Wikipedia articles. We will use the Google Natural Questions dataset, along with which an excellent visualization helpful for understanding the idea behind QA can be found at https://ai.google.com/research/NaturalQuestions/visualization.
The basic idea can be summarized as follows: for each article-question pair, you must predict/select long- and short-form answers to the question drawn directly from the article:
- A long answer would be a longer section of text that answers the question—several sentences or a paragraph.
- A short answer might be a sentence or phrase, or even in some cases a simple YES/NO. The short answers are always...
