시퀀스-투-시퀀스를 활용한 데이터 기반 문장 생성 모델 비교

Abstract

In this study, we compare the modified encoding method and attention method using a sequence-to-sequence model in the Data2Text task. In addition, we propose a model that complements the shortcomings. Data2Text refers to the task of generating description from tabular data. The data consists of a set of values and fields, and there are significant performance differences depending on how the model learns these structures. Therefore, we compare two representative models utilizing sequence-to-sequence under the same conditions to find a more effective methodology for learning the structure of the data. In addition, it adds a copy mechanism to improve performance by allowing the output of words that are not in the word vocabulary, such as proper nouns. As a result of the experiment, encoding methods using values, fields, position information and dual attention, which combines values and field attention, showed the best performance. In addition, performance improvement in the sequence-to-sequence model was confirmed when the copy mechanism was utilized.