An Extractive Closed domain Question Answering based on the domain comprising of Sebi Documents.
Deepti Saravanan, Sandeep Dash, Vrinda Lakhotia, Pulkit Parikh, Lini Thomas, Kamal Karlapalem
SEBI has a vast amount of regulation database, which is tough to stir through or have perfect command over all the information, and it is also too time-consuming and hard to provide answers to user queries, open to human errors. Handling extensive information and providing confident answers from it are expensive tasks in time and resources. Creating query-based visualizations is also a hectic activity, considered the number of queries. Hence, our project tries to automate it, with higher confidence in the answers.
The primary purpose of the project is to provide a pipeline that returns answers to the user's queries, with relevant visualizations and texts from the SEBI documents, i.e., given a Query and Document(s), to provide an Answer span, Answer regulation, and supporting interactive visualizations using Knowledge Graph. The project aims to do this with the employment of four components: Query preprocessing, Regulation Retrieval, Answer module, and Semantic Knowledge graph module.
Data: Here
Live Demo: Here
Reference Document: Here