Original scientific paper

https://doi.org/10.17559/TV-20250224002415

Entity Type-Aware Ultra-Fine Entity Typing with Adaptive Distance Optimization

Feng Wang ; School of Computer Science and Technology, ZhouKou Normal University, ZhouKou 466001, China *
YiXiu Qin ; School of Computer Science and Technology, ZhouKou Normal University, ZhouKou 466001, China

* Corresponding author.

Abstract

Recent work have focused on representing entity mentions as well as entity types in a high-dimensional box space to more effectively capture their complex relationships. In box space, entity mentions as well as entity types are represented as high-dimensional hyper-rectangles. However, the role of entity types is often overlooked or inadequately incorporated in classification tasks within box space. Furthermore, the model struggles with precise optimization under specific conditions, or its constraints are excessively strict, leading to the complete overlap of the two box center points. In light of these shortcomings, this paper presents the Entity Type-aware Ultra-fine Entity Typing with Distance Optimization. The ETUT effectively integrates entity type information and introduces an adaptive distance-based module, ensuring model optimization even when two boxes are either fully separated or completely overlapped in space. Experimental results from ultra-fine and fine-grained entity typing datasets demonstrate the effectiveness of ETUT, showing it to be a state-of-the-art method in the domain of ultra-fine and fine-grained entity typing.

Keywords

fine-grained entity typing; multi-objective optimization; type hierarchy; ultra-fine entity typing

Hrčak ID:

332857

URI

https://hrcak.srce.hr/332857

Publication date:

29.6.2025.

Visits: 213 *