Roland Schäfer, Adrien Barbaresi & Felix Bildhauer (2013) The Good, the Bad, and the Hazy: Design Decisions in Web Corpus Construction. In Proceedings of the 8th Web as Corpus Workshop (WAC-8). [BibTeX]
In this paper, we examine notions of text quality in the context of web corpus construction. Web documents often contain material which disqualifies them from inclusion in a corpus (tag clouds, lists of names or nouns, etc.). First, we look at the agreement between coders (especially corpus designers) given the task of rating text quality. Then, we evaluate a simple and fully unsupervised method of text quality assessment based on short and very frequent words. Finally, we describe our general approach to the construction of carefully cleansed and non-destructively normalized web corpora. Under this approach, we annotate documents with quality metrics instead of actually removing those documents classified as being of low quality.