There are many possible methods of text conversion and one such technique is screen-scraping. The primary use of screen-scraping software is migrating information from one computer system onto another by reading data from specific fields in the legacy system’s user interface and making the data available as objects.
It is possible to apply screen-scraping to Web documents, however its disadvantages make it useful only under specific circumstances. Screen-scraping techniques identify specific sections within single documents and extract the data contained within each section.
This is useful when the documents’ structure and layout are uniform and formally controlled, i.e. the sections are structured very clearly in a table, for example, and the layout of the table is controlled by a content management system that ensures each page is presented in a uniform style.
Cases where this would be of use are; extracting article headlines from a single newspaper’s homepage, or extracting book titles from individual item pages on a Web based bookstore, like Amazon.co.uk, because each item page is presented in a controlled uniform layout. However, the same techniques will not be portable onto another newspaper’s homepage where the headline sections may be structured differently and nor will they be applicable to any other online bookstore because the layout of item descriptions will not be the same as Amazon’s style.
This makes screen-scraping software portability more difficult because it necessitates modification for use with any documents other than the original design. Screen-scraping is also a high maintenance approach because it requires modification to cope with any changes to the original documents; if Amazon changes their presentation style then screen-scrapers targeting the original interface will have to be updated.
Although the content of academic staff homepages appears to be similar, and therefore must be governed by departmental guidelines, nevertheless it is empirically evident that their structure and layout is not uniform or formally controlled in the same way as the Web pages of their home institution. It is clear then that the domain of your project will affect your choice to implement a screen-scraping approach.
In addition, screen-scraping software does not include tasks such as co-reference recognition (CORC) that involves more sophisticated programming techniques, which make use of Artificial Intelligence (AI) practices to analyse document content and extract relevant information. CORC determines whether a piece of data refers to an item that has already been extracted. For example, screen-scraping software would not be able to determine that a professor named on his homepage is the same professor named on the homepage of a project that he is involved with. Such AI techniques are covered by Information Extraction (IE) that includes two traditions; Natural Language Processing (NLP) and Wrapper Induction (WI).
Example NLP approaches; GATE and Monrai Cypher.
Example WI approaches; WIEN, Stalker, RoadRunner, and BWI
No comments:
Post a Comment