method: QA_Base_MRC_12020-05-15

Authors: SIJIN WU, QIMING PENG, YUDI CHEN, JIANJIAN CAO, YOUHUI GUO, GANGYAN ZENG

Description: Our method is based on Ernie, Albert, and unilm. The content text of the Document is first encoded by the above-mentioned models and then decoded by our MRC method to predict the answers. A weighted voting strategy is used to merge the answers of different models to the final results.