Developing a Dataset of Overridden Information in Wikipedia

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5601–5608 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5601 Developing a Dataset of Overridden Information in Wikipedia Masatoshi Tsuchiya, Yasutaka Yokoi Toyohashi University of Technology 1–1 Hibarigaoka, Tempaku-cho, Toyohashi, Aichi, Japan {tsuchiya, yokoi}@is.cs.tut.ac.jp Abstract This paper proposes a new task of detecting information override. Since all information on the Web is not updated in a timely manner, the necessity is created for information that is overridden by another information source to be discarded. The task is formalized as a binary classification problem to determine whether a reference sentence has overridden a target sentence. In investigating this task, this paper describes a construction procedure for the dataset of overridden information by collecting sentence pairs from the difference between two versions of Wikipedia. Our developing dataset shows that the old version of Wikipedia contains much overridden information and that the detection of information override is necessary. Keywords: Information Override; Recognizing Textual Entailment 1. Introduction The vast amount of information provided through the Web has become an indispensable tool in our daily lives. The appropriate selection of this infor- mation is necessary because it is unrealistic to re- fer to all information (Marshall and Shipman, 1997; Macskassy and Provost, 2001). In particular, since all information on the Web is not updated in a timely man- ner, it is essential to discard information overridden by another information source. As an example, consider the two sentences shown in Table 1. Since Oshida Sta- tion was discontinued in 2016, as described in the sen- tence sr, it is inappropriate to refer to the sentence st for the transportation method to Asagishi Aza Oshida, Morioka City, in 2022. The Web contains much outdated information like the sentence st, mation utilization.which is an obstacle to infor- To support the use of such information, methods have been proposed to or- ganize and present this information sequentially (Fung et al., 2007; Wang et al., 2008). These studies, however, do not provide a way to automatically deter- mine whether a particular piece of information is inap- propriate for current reference from a users’ perspec- tives. Based on the above discussion, we propose a new task detecting overridden information like the sentence st. There are two possible settings for detecting overridden information: one is to determine whether the reference sentence sr has overridden the target sentence st using both the target sentence st and the reference sentence sr as input, and the other is to determine whether the target sentence st has been overridden using only the target sentence st as input. Considering the original goal, which is to support information usage by auto- matically detecting overridden information, the latter setting, which takes only the target sentence as input, is more reasonable. However, to realize the latter setting, it is necessary to solve the subtask to discover the ref- erence sentence that overrides the target sentence at the st As of the year …

Content truncated to 500 words