IBM Watson™ Discovery Service Ideas

We've moved...

You'll be redirected shortly, we've moved to our new idea portal:

Support creation of custom table labels in SDU that return content in a cell tagged with a unique row/column value

We have many use cases where customers want to be able to analyze tabular data. For example, there are many cases where a customer wants to analyze financial reports to understand how a certain company has been performing. Much of this data is presented in tables in a PDF. To do this, we need to be able to identify not only the cell, but also its context. For example, we might want to know what a company's revenue was in a certain year.

Right now we can annotate tables as tables, and identify the table row and column headers, but we want to take the next step and identify the name of the column and rows and give each bit of text in a cell a label or tag that explicitly identifies it. This has enormous potential in terms of business value and product differentiation.   

  • Guest
  • May 17 2019
  • Future Consideration
Why is it useful?
Who would benefit from this IDEA? As a customer I want to be able to include tabular data in my analysis, or I want to be able to answer questions with specific information form a cell in a table.
How should it work?
Idea Priority
Priority Justification
Customer Name
Submitting Organization
Submitter Tags
  • Attach files
  • MARK RICE commented
    June 21, 2019 22:36

    Here is an example of what I mean...

    Attachment "sample Table.png" shows a typical example of tabular data set of data in a table that some clients want to be able to extract. In this case, these are analysis results of ground soil for a proposed golf course. The client wants to extract the data from each row as 4 pieces of data, with the data in each row related.

    The way we do this now is to write parsing rules on the text file derived from that pdf. Attachment "data from sample table in CA Studio with correct annotations.pdf" shows how this looks in CA Studio after the rules have been successfully built. Each pink highlighted row is essentially a set of 4 measurements all related to the first column. The yellow box shows the main annotation (the covered text), with the various features (in this case the features are the related measurements). 

    So we have been successful here, but it took a few hours to design the rules and a couple of days to build them, test them, refine them and test them again. If we could do this mapping and extraction in SDU, we could have the training done in minutes. In addition, it is likely SDU will be more accurate, as the rules apporahc asumes all tables are rendered the same way in the text file after the OCR. This is not always the case with pdf documents.