A BiLSTM-CRF Entity Type Tagger for Question Answering System
Cheng-Yun Kuo, Eric Jui‐Lin Lu · 2021
Question answering system over linked data (QALD) has been a very important research field in natural language processing (NLP). And the process of detecting useful words and assigning them with right entity types is crucial to the performance of QALD systems. Although entity-type taggers achieved good results using probability graph models such as MEMM and CRF, the design and selection of features may pose limitations. Due to the popularity of deep learning architectures, many studies employed Recurrent Neural Network (RNN) framework and achieved state-of-art performances in NLP. Therefore, we choose to use BiLSTM-CRF in the design of entity-type tagger. It can be seen from the experimental results that the proposed BiLSTM-CRF model outperformed other probability graph models, which also lead to the best performance of overall Question Answering system than other competitor systems.