EURO 2025 Leeds
Abstract Submission

1999. Contrastive Learning-based Semantic Embedding of Technology Groups

Invited abstract in session TD-38: Foundation Models and Optimization, stream Data Science meets Optimization.

Tuesday, 14:30-16:00
Room: Michael Sadler LG19

Authors (first author is the speaker)

1. Yoon SUA
Industrial management, Myongji University
2. Dohyun (Norman) Kim
Myongji University

Abstract

Accurate embedding vectors that effectively capture the semantics of technology groups are essential in technology analytics. Traditional approaches commonly derive these embeddings by averaging individual paper embeddings within a technology group. However, this simple averaging often fails to preserve distinctive semantic features, diluting critical contextual information unique to each technology domain. To overcome this limitation, we propose a novel embedding approach using contrastive learning between technology groups and their associated paper embeddings. Our method explicitly models semantic relationships, resulting in embeddings that accurately reflect the inherent meaning and contextual differences among technology groups. Experimental evaluations demonstrate that our contrastive learning-based embeddings significantly outperform conventional averaging methods. Our results suggest that the proposed embedding methodology is expected to be effective and practically useful in diverse technology analysis tasks.

Keywords

Status: accepted


Back to the list of papers