Open AccessAn Empirical Study of Information Retrieval and Machine Reading Comprehension Algorithms for an Online Education Platform
Author(s) -
Eduardo Fernandes Montesuma,
Lucas C. Carneiro,
Adson Damasceno,
João Victor F. T. de Sampaio,
Romulo F. Férrer Filho,
Paulo Henrique M. Maia,
Francisco Oliveira
Publication year - 2021
Language(s) - English
Resource type - Conference proceedings
DOI - 10.5753/stil.2021.17801
Subject(s) - computer science , pipeline (software) , context (archaeology) , reading (process) , information retrieval , artificial intelligence , question answering , empirical research , reading comprehension , natural language processing , comprehension , machine learning , paleontology , philosophy , epistemology , political science , law , biology , programming language
This paper provides an empirical study of various techniques for information retrieval and machine reading comprehension in the context of an online education platform. More specifically, our application deals with answering conceptual students questions on technology courses. To that end we explore a pipeline consisting of a document retriever and a document reader. We find that using TF-IDF document representations for retrieving documents and RoBERTa deep learning model for reading documents and answering questions yields the best performance with respect to F-Score. In overall, without a fine-tuning step, deep learning models have a significant performance gap with comparison to previously reported F-scores on other datasets.