BAS: An Answer Selection Method Using BERT Language Model

  • 2019-11-11 19:31:09
  • Jamshid Mozafari, Afsaneh Fatemi, Mohammad Ali Nematbakhsh
  • 0

Abstract

In recent years, Question Answering systems have become more popular andwidely used by users. Despite the increasing popularity of these systems, thetheir performance is not even sufficient for textual data and requires furtherresearch. These systems consist of several parts that one of them is the AnswerSelection component. This component detects the most relevant answer from alist of candidate answers. The methods presented in previous researches haveattempted to provide an independent model to undertake the answer-selectiontask. An independent model cannot comprehend the syntactic and semanticfeatures of questions and answers with a small training dataset. To fill thisgap, language models can be employed in implementing the answer selection part.This action enables the model to have a better understanding of the language inorder to understand questions and answers better than previous works. In thisresearch, we will present the "BAS" (BERT Answer Selection) that uses the BERTlanguage model to comprehend language. The empirical results of applying themodel on the TrecQA Raw, TrecQA Clean, and WikiQA datasets demonstrate thatusing a robust language model such as BERT can enhance the performance. Using amore robust classifier also enhances the effect of the language model on theanswer selection component. The results demonstrate that language comprehensionis an essential requirement in natural language processing tasks such asanswer-selection.

 

Quick Read (beta)

loading the full paper ...