<![CDATA[Statistical metadata plays an important role in society. Statistical metadata allows us to know all information about statistical activities that have been carried out. In this study, we built a Closed Domain Question Answering system related to statistical metadata (CDQA-Metadata Statistik). The absence of a large benchmark regarding QA datasets on statistical metadata caused us to choose the transfer learning method. This study uses a retriever (BM25)-reader (IndoBERT) architecture based on transfer learning with three experiments. The results of the first experiment showed that statistically the performance of the transfer learning model significantly outperformed the non-transfer learning model on human question data and automatic question data. The results of the second experiment showed that statistically the performance of the CDQAStatistical Metadata system based on transfer learning on automatic question data was significantly better than on human question data. The results of the third experiment showed that for human question data, adding automatic question data during fine-tuning did not improve system performance. Then on automatic question data, adding human question data during fine-tuning did not seem to be able to improve system performance.]]>
Copyrights © 2024
