Automated Keyword Generation Using LLMs in a High-Volume Data Virtualization Service

概要

In modern data infrastructure, a data virtualization service (DVS) allows users to create many versions of source datasets for different processing tasks. For example, a multivariate or multi-modality dataset can be used for different machine learning tasks (e.g., classification, detection, segmentation, forecasting, and so on). Data selection, participation, integration, and preprocessing in different tasks result in many virtual datasets [Khan et al., SSC 2026]. Meanwhile, different users need to discover existing virtual datasets frequently and easily to avoid duplicate efforts. To enable an effective search engine in a DVS, it is desirable to automatically generate keywords for each virtual dataset. In this work, we present a solution for generating keywords for virtual datasets using large language models (LLMs). We propose five strategies for selecting different collections of text data for processing by the LLMs. We applied the five strategies to virtual datasets created in five application case studies with different source datasets. We conducted experiments with three LLM generators and two embedding techniques. The experiments show that the performances of the five strategies depend on the quality of the text descriptions. The results help us adopt a multi-strategy implementation of a new keyword generation service (KGS) as part of an extended DVS. The work demonstrates that the capability of automated keyword generation is indispensable for a high-volume DVS that is dataset-rich and frequently accessed.

参考リンク

• Paper ↗

リンク先のライセンス・再現性・データ条件をご確認ください。自動分類や実装の目安は保証ではありません。

実装を始めるプロンプト

参考リンクを確認したうえで、次の内容をコピーして利用できます。

# 論文・記事
Automated Keyword Generation Using LLMs in a High-Volume Data Virtualization Service

# 概要
In modern data infrastructure, a data virtualization service (DVS) allows users to create many versions of source datasets for different processing tasks. For example, a multivariate or…

# 原典
https://arxiv.org/abs/2610.41893

上記の手法を実装してください。実行前に依存関係とライセンスを確認してください。