This document-based question answering application is based on an AI (Artificial Intelligence) algorithm that automatically answers questions based on facts from given articles. Given an article where the answer to a question can be found as a segment of text, the AI engine automatically locates the answer that is span from the article. The technology owner has embedded a state-of-the-art deep learning-based question answering engine at the backend to perform the answer extraction task.
The proposed technology consists of a series of question answering solutions and API services, including but not limited to:
- Automated answer extraction algorithm for articles of any topic.
- API services for question answering in web.
- Automatically locating key information across long passages.
- Automatically capturing relations between entities in an article.
The current state of the technology supports English language question answering solution.
The proposed technology can be applied in building personal assistants such as chatbots that answer users’ questions about a product, a service or other topics, where the answers can be extracted from resources that are text based, such as existing FAQ databases, existing web pages, news articles, email archives, etc.
Existing systems usually retrieve complete sentences or paragraphs as answers to a user’s question. For questions whose answers are more specific such as a date, a location or an organization’s name, existing systems may not be able to accurately locate the answers. This system is trained to specifically answer questions whose answers are short spans of text.