BlogSet-BR: A Brazilian Portuguese Blog Corpus

BlogSet-BR: A Brazilian Portuguese Blog Corpus 1Henrique D.P. dos Santos, 2Vinicius Woloszyn, 1Renata Vieira 1Pontificia Universidade Catolica do Rio Grande do Sul Av. Ipiranga, 6681, Building 32 - Porto Alegre - Brazil 2Instituto de Inform´atica – Universidade Federal do Rio Grande do Sul Caixa Postal 15.064 – 91.501-970 – Porto Alegre – RS – Brazil henrique.santos.003@acad.pucrs.br, vwoloszyn@inf.ufrgs.br, renata.vieira@pucrs.br Abstract The rich user-generated content found on internet blogs have always attracted the interest of scientific communities for many different purposes, such as from opinion and sentiment mining, information extraction or topic discovery. Nonetheless, an extensive corpora is essential to perform most of Natural Language Processing involved in these tasks. This paper presents BlogSet-BR, an extensive Brazilian Portuguese corpus containing 2.1 billions words extracted from 7.4 millions posts over 808 thousand different Brazilian blogs. Additionally, a survey was conducted with authors to draw a profile of Brazilian bloggers. Keywords: Blog as Corpus, BlogSet-BR, Brazilian Portuguese Corpus 1. Introduction Several efforts have been made to build a large corpora based on user-generated content, since they are crucial for many different Natural Language Processing tasks, such as opinion mining, sentiment analysis, topic detection and Age/Gender detection (Burton et al., 2009; Buck et al., 2014; Santos et al., 2017). From all content available on the internet, blogs have been often employed as a main source of user-generated content (Agarwal and Liu, 2008; Agar- wal and Liu, 2009; Santos et al., 2012). Nonetheless, there is still a lack of a large semi-structured corpus that also con- tains author profiles in Brazilian Portuguese. To illustrate this situation, Table 11 gives an overview of the initiatives to create a Portuguese corpora based on general content. Besides Linguateca resources in Table 1, we added other four corpora: the brWac corpus for Brazilian Portuguese built by downloading text from the web (Boos et al., 2014), Buscap´e, a Portuguese product reviews corpus extracted from a collaborative review site (Hartmann et al., 2014), the Portuguese Wikipedia dump and BlogSet-BR corpus. The first three corpora listed in the table contains the authorship of the content. This paper describes BlogSet-BR, a semi-structured large corpus containing author information extracted from Brazilian Portuguese blogs. It contains more than 7.4 mil- lions posts resulting in 2.1 billion words. It is the first Brazilian Portuguese corpus about blogs. In this paper, we also conducted a survey with the authors to create a profile of Brazilian bloggers. The main contributions of this work are: • The first semi-structured large corpus of posts from Brazilian Portuguese blogs; • A corpus with authorship, date-time and label infor- mation attached to the text; 1Adapted from http://www.linguateca.pt/ACDC/ • Information retrieved from the author profiles, such as gender, age, and educational level. This work describes the building process of the BlogSet- BR collection and also descriptive statistics about the data. The rest of this paper is organized as follow: first, Sec- tion 2. presents related work; in Section 3., we discuss the three phases of creating the BlogSet-BR collection; section 4. provides an overview …

Content truncated to 500 words