Entity-Aspect Linking Dataset

Here you can find all resources used in the paper (link-to-be-added) “EAL: A Toolkit and Dataset for Entity-Aspect Linking”, which I wrote together with Jingyi Zhang, Ferdinand Betz and Kiril Gashteovski; it will be presented at JCDL 2019, this June. The work is an extension of a previous paper, which I wrote together with Simone Paolo Ponzetto and Laura Dietz and presented at JCDL 2018.


We present both the initial sentences+triples extracted from OPIEC and the related EAL dataset (with removed duplicates and further filtering, as described in our paper).

The OPIEC output has one sentence+triple per line, reporting the following information:

Sentence:  Triple:  factuality: quantities:  attribution:  time: space: Links: (SubjLink: ObjLink:)

The EAL-D is a JSON file structured as follows:

{id_context_1: {“entity_mention”: {“entity”:entity_id, “mention”:mention}, “true”:correct_aspect_id,”sent_context”: sentence,  “aspect_candidates” : [{“id_aspect”: id_aspect, “content”: content, “header”: header, “entities”:entities}, …], id_context_2: … }

Over 8K sentences with entity-aspect link on subject of the relation [OPIEC, EAL-D]

Over 13K sentences with entity-aspect link on object of the relation [OPIEC, EAL-D]

Over 1K sentences with entity-aspect link on both elements of the relation [OPIEC, EAL-D]


Check out our Entity-Aspect Linking demo!