Design Challenges in Wikipedia-style Spotlight Annotation. Issue 1 (April 2021)

• Record Type: Journal Article
• Authors: Pang, Kunyuan; Tang, Jintao; Li, Xinyi; Li, Xiao; Li, Shasha; Wang, Ting
• Is Part Of: Journal of physics. Volume 1871:Issue 1(2021)
• Publication Date: 2021-04
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1871/1/012096
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Wikipedia has set up a manual of style to provide the Wikipedian with guidelines to create links as spotlights in articles to help readers' understanding. Wikification is coined following Wikipedia's style guide as a task of automatically detecting mentions and linking to entities in plain-text articles. However, existing methods rely heavily on the mention-entity dictionary and focus on in-map entity precision. Some popular methods are limited to Named Entities. The comparison on the outcome of spotlight annotation is insufficient. We investigate the design challenges for constructing automatic annotating in Wikipedia-style systems. A dataset is built simulating real-world spotlight annotation problems in both the preparing and application phases. Popular methods are compared along with a straightforward framework we propose called WikiER. Experiments show that the performance of currently available systems varies a lot. It is important to train on the task-specific data. The best performance is achieved by WikiER.