Extração de dados na web: transformando listas HTML em formato tabular

Abstract

Some HTML constructs are used not only for visualization purposes but also to structure information. One of such structures is used to represent lists. Looking at the markup of HTML lists, it is possible to see that the information is organized as a collection of records. Given this, some works aim at transforming the text records into tables, composed by records and columns. One of the most known works, called ListExtractor, depends on existent knowledge bases in order to perform the transformation. This kind of data source structuring is useful in researches related to dataspaces, and have direct application in areas like data integration, knowledge extraction and information retrieval. In this context, the goal of this work is creation a series of statistical rules whose goal is to transform lists into tables. Generally speaking, the rules explore the presence and frequency of special characters inside the text to perform the contente segmentation, and they do not rely on existent knowledge bases. The proposed rules correspond to data extractors, which can be incorporated in dataspaces architectures. Experiments show the extractors performance when applied to real HTML lists found on the Web. Additionally, a comparison is made between the extractors and the ListExtractor approach.