Headword-Oriented Entity Linking: A Special Entity Linking Task with Dataset and Baseline

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1910–1917 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1910 Headword-Oriented Entity Linking: A Special Entity Linking Task with Dataset and Baseline Mu Yang†, Chi-Yen Chen†, Yi-Hui Lee†, Qian-Hui Zeng†, Wei-Yun Ma∗†, Chen-Yang Shih‡, Wei-Jhih Chen‡ †Institute of Information Science, Academia Sinica, Taipei, Taiwan ‡R&D Center, PIXNET Corporation, Taipei, Taiwan emfomy@gmail.com, chiyench@usc.edu, yi-hui.lee@utdallas.edu, littlemiss330@gmail.com, ma@iis.sinica.edu.tw, kent@pixnet.tw, wayne@pixnet.tw Abstract In this paper, we design headword-oriented entity linking (HEL), a specialized entity linking problem in which only the headwords of the entities are to be linked to knowledge bases; mention scopes of the entities do not need to be identified in the problem setting. This special task is motivated by the fact that in many articles referring to specific products, the complete full product names are rarely written; instead, they are often abbreviated to shorter, irregular versions or even just to their headwords, which are usually their product types, such as “stick” or “mask” in a cosmetic context. To fully design the special task, we construct a labeled cosmetic corpus as a public benchmark for this problem, and propose a product embedding model to address the task, where each product corresponds to a dense representation to encode the different information on products and their context jointly. Besides, to increase training data, we propose a special transfer learning framework in which distant supervision with heuristic patterns is first utilized, followed by supervised learning using a small amount of manually labeled data. The experimental results show that our model provides a strong benchmark performance on the special task. Keywords: Corpus, Information Extraction, Distant Supervision 1. Introduction Named entity linking, or entity linking (EL), is determin- ing the identity of entities that are mentioned in the text and bridging them from unstructured textual data with struc- tural knowledge bases. Traditionally, entity linking in- volves two sequential tasks: first, detecting mentions using named entity recognition, seeking to locate entity chunks in text and classify their entity types; second, linking the recognized entity mention with the corresponding entry in a knowledge base. Domain-specific entity linking has attracted attention due to commercial demands in practice, especially product name linking. Unlike conventional named entities such as per- son, location, or organization names, complete full product names are rarely written in context; instead, they are often abbreviated to a shorter, irregular version or even just to their headwords. For instance, in a real blog article about “Dior Addict Lacquer Stick” (PID 17755 in our cosmetic database) and other cosmetic products, the ‘PID 17755’ item is sometimes written as “Dior lacquer stick”, or just “pen-shaped lacquer stick” or even abbreviated to its prod- uct type — “stick”, such as “I really like the stick that Dior released last month,” where “stick” refers to ‘PID 17755’. In most cases, the mentions contain at least the product type to represent the product. Such being the case, from a prac- tical application perspective such as product …

Content truncated to 500 words