Creating a DEXTL Specification from the Tags¶
After the structure is created and the tokens of the page are known, we are ready to write the DEXTL specification to extract messages. Click the “Edit” button of the Generation pane; a warning dialog will appear, because the default specification uses the “<Extractor_name>_ITEM” name for the root level, but the structure was renamed to WEBMAIL_MANUAL. When clicking “Ok”, you can see that the specification skeleton is changed to reflect the new name of the root level that was specified in the Structure pane.
To write the specification, we are going to replace the pattern that appears in the skeleton (“Replace this with Pattern Components”), with the following:
The check box and the index of the message are not going to be extracted; for the purposes of this example, we will ignore them.
The first item to extract is the date of the message, with structure TD “31/01/2007” ENDTD (from the Token Viewer). We want to extract it into the MESSAGEDATE field, so we add the following pattern:
TD :MESSAGEDATE ENDTD
That will extract the text inside a TD into the MESSAGEDATE field.
The next item is the sender of the e-mail, with pattern TD ANCHOR “John Smith” ENDANCHOR ENDTD; in a similar way, the pattern that will extract this text is
TD ANCHOR :SENDER ENDANCHOR ENDTD
The pattern for the subject (TD ANCHOR “W3C Holds Workshop on Frameworks for Semantics in Web Services : 2005-06-09” ENDANCHOR ENDTD ) is very similar to the sender one
TD ANCHOR :SUBJECT ENDANCHOR ENDTD
Finally, the pattern for the size of the e-mail is almost the same than for the date:
TD :SIZE ENDTD
In summary, putting all the sub patterns together, the pattern to extract a message is
TD :MESSAGEDATE ENDTD TD ANCHOR :SENDER ENDANCHOR ENDTD TD ANCHOR
:SUBJECT ENDANCHOR ENDTD TD :SIZE ENDTD
The final specification is shown in DEXTL specification to extract e-mail data; click the “Save” button to accept it. Testing it will retrieve the same results that were extracted in Part I of the webmail example.
Please read the ITPilot DEXTL manual (ITPilot DEXTL Guide) for more information about the syntax of the specifications.
Generating FROM/TO Patterns¶
It is sometimes useful to explicitly determine the part of the page at which the information extraction process is to begin and the part at which it is to end. A typical example is shown below. Tabulated Results of a BookshopResult of the DEXTL Program Test on DETAIL shows the graphic format used by an on-line bookshop to show information on its products.
A DEXTL program generated carelessly could return the heading row as yet another result, because it might sport the same internal HTML structure than the table contents.
When DEXTL programs are generated from examples, most of these situations are either transparently handled by ITPilot or can be solved using the Disambiguate option. Nevertheless, it is also possible to explicitly limit the part of a document, where matches will be searched. To do this, follow these steps:
If you have not done it yet, launch a browser using the Browser > New Browser menu option and browse to the target page
Select the text delimiting the beginning of the area in the browser (the table heading in the example of Tabulated Results of a BookshopResult of the DEXTL Program Test on DETAIL).
Return to the Extractor wizard and click on . The system will extend the DEXTL specification to include the search boundary. You can also include a boundary for the end of the search zone by using the button.