[3K5-OS-2b] OS-2

[3K5-OS-2b-04] Fine-Tuned Data Representation Models for Data Exploration in Data Markets

〇Kosuke Manabe¹, Yukihisa Fujita², Masahiro Kuwahara², Teruaki Hayashi¹ (1. The University of Tokyo, 2. Toyota Motor Corporation)

Keywords:Language Models, Contrastive Learning, Data Market

With the development of information technology for data collection, storage, and analysis, data collaboration and utilization in different fields are attracting attention. In this climate, data markets are emerging to exchange data across fields. However, data analysis experience and data format expertise are needed to explore and discover the data related to our interests in the data exchange platforms. In addition, current metadata is mainly created manually, and the consistency and interpretability of descriptions are highly dependent on the knowledge and ex- perience of the data providers. To address the above issues, we propose a method for learning a data representation model that takes data bodies as input and outputs embedded representations for data retrieval. As a result, we found that the proposed method can obtain a data representation that more accurately reflects the topic of the data than existing methods.