Multi-class Multilingual Classification of Wikipedia Articles Using Extended Named Entity Tag Set

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1197–1201 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1197 Multi-class Multilingual Classification of Wikipedia Articles Using Extended Named Entity Tag Set Hassan S. Shavarani*, Satoshi Sekine Simon Fraser University, AIP Center for Advanced Intelligence BC, Canada, Riken, Tokyo, Japan sshavara@sfu.ca, satoshi.sekine@riken.jp Abstract Wikipedia is a great source of general world knowledge which can guide NLP models better understand their motivation to make predictions. Structuring Wikipedia is the initial step towards this goal which can facilitate fine-grain classification of articles. In this work, we introduce the Shinra 5-Language Categorization Dataset (SHINRA-5LDS), a large multi-lingual and multi-labeled set of annotated Wikipedia articles in Japanese, English, French, German, and Farsi using Extended Named Entity (ENE) tag set. We evaluate the dataset using the best models provided for ENE label set classification and show that the currently available classification models struggle with large datasets using fine-grained tag sets. Keywords: SHINRA-5LDS, Multi-lingual Multi-labeled Classification, Extended Named Entity 1. Introduction Major progress has been made in different tasks in Natural Language Processing, yet our models are still not able to describe why they make their decisions when summarizing an article, translating a sentence, or answering a question. Lack of meta information (e.g. general world knowledge regarding the task) is one important obstacle in the construction of language understanding models capable of reasoning about their considerations when making decisions (predictions). Wikipedia is a great resource of world knowledge for human beings, but lacks the proper structure to be useful for the models. To address this issue and make a more structured knowledge-base, Sekine et al. (2018b) try to structure Wikipedia. Their final goal is to have, for each Wikipedia article, known entities and sets of attributes, with each attribute linking to other entities wherever possible. The initial step to- wards this goal would be to classify the entities into predefined categories and verify the results using hu- man annotators1. Throughout the past years, many have tried classi- fying Wikipedia articles into different category sets mostly containing between 3 to 15 class types (Toral and Munoz, 2006; Watanabe et al., 2007; Dakka and Cucerzan, 2008; Chang et al., 2009; Tardif et al., * The author was an intern at AIP Center for Advanced Intelligence, during this project. 1 Please note that the verification process plays an im- portant role in the knowledge-base construction process since it leads to what is represented to our models as world facts. 2009). Such categorization type sets are not much helpful when the classified articles are being used as the training data for question answering systems, since the extracted knowledge-base does not provide de- tailed enough information to the model. On the other hand, much larger categorization type sets such as Cyc-Taxonomy (Lenat, 1995), Yago- Taxonomy (Suchanek et al., 2007), or Wikipedia’s own taxonomy of categories (Sch¨onhofen, 2009) are not suitable for classifying Wikipedia articles since the tags are not verifiable for annotators2. In …

Content truncated to 500 words