|
Activity Number:
|
256
- Contributed Poster Presentations: Section on Statistical Learning and Data Science
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 29, 2019 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #306405
|
|
|
Title:
|
Question Answering Using a Domain Specific Knowledge Base
|
|
Author(s):
|
Mitchell Kinney*
|
|
Companies:
|
University of Minnesota - Twin Cities
|
|
Keywords:
|
Question Answering;
Knowledge Base;
Graph Neural Network;
Negative Sampling
|
|
Abstract:
|
Using recent work from Battaglia et al. on building the framework for graph neural networks, I propose to construct a domain specific knowledge base to perform challenging question answering. From Allen Institute's A12 Reasoning Challenge (ARC), challenging question answering is defined as answering multiple choice questions based on a large (~14 million sentences) collection of information that cannot be answered using traditional methods. The novelty of this method lies in using negative sampling as the objective function with BERT sentence embeddings which has not been combined yet for the purpose of question answering. Future work includes making attention mechanisms for word importance included so that the process is end-to-end differentiable.
|
Authors who are presenting talks have a * after their name.
