Polibits, Vol. 45, pp. 13-19, 2012.
Abstract: Relational data is often encoded in tables. Tables are easy to read by humans, but difficult to interpret automatically. In cases where table layout cues are not obtainable (missing HTML tags) or where columns are distorted (by copying from a spreadsheet to text) previous table extraction approaches run into problems. This paper introduces a novel table parsing approach. Our approach is based on a set of simple assumptions: (a) every table can be split up in data cells and headers, and (b) every table can be parsed beginning from a data cell utilizing the overall table structure. The table parsing is defined as ``table flattening'' in this paper. That is, the parsing starts with a data cell and pulls out all token (i.e., headers and sub-headers) associated with a respective data cell. We propose a parsing technique that uses two simple parsing heuristics: table headers are to the left of and above a data cell. We experimented with trader emails that contained instrument information with bid-ask prices as data cells. We developed a clustering and classifying method for finding prices reliably in the data set we used. This method is transferable to other data cell types and can be applied to other table content.
Keywords: Information retrieval, document processing, tables
PDF: A Flexible Table Parsing Approach
PDF: A Flexible Table Parsing Approach