Text Summarization Tool:
- About
- This is an automatic
text summarization application, it extracts
the most pertinent sentences from a given article and attempts to
minimize redundancy. There are two main ways to use the application,
either via a simple cut and paste dialog or if you wish, by loading a
text file with the ".txt" extension.
- A typical use case might
involve speed reading of a new technical article without an abstract.
After launching the application, you will be presented with a window
that is made up of four main elements. At the very top are three tabs,
labeled "Ad hoc text", "Text from Files" and
"Results" in that order. Below the tabs, one will find a
white text box, where the article to be summarized may be pasted into.
Under the white box are three buttons, "Demo",
"Summarize" and "Clear". These are pretty self
explanatory. Finally, one comes to the bottom of the window, where
there is a sliding scale. This slider determines the number of
sentences that are going to be extracted from the article. Just click
on the little trapezoid and drag to the required number, release your
mouse button and click on "Summarize" to begin the process.
Once the summary is created, the application will automatically display
the results and switch the view to the third tab, labeled
"Results".
- The "Demo"
button summarizes Lincoln's Gettysburg
address into two lines.
- The summarization
uses a matrix decomposition to determine the best clusters of sentences
from the article and chooses the best representative from each cluster
to include in the summary. The choice is biased towards earlier
sentences and sentences of intermediate length, but otherwise, the
choice is solely due to the authority of the sentence given the terms
it contains (similar to the SALSA algorithm).
Downloads:
Screenshot:
Acknowledgments
- This tool requires JVM
1.5 and above to run. The underlying summarization engine arose from a
project done for the Text Processing on the Web Class by Dr Kan Min Yen
in the Spring of 2007. The GUI was implemented
to ease the use of the tool.
- The libraries used in
the application are included in the distribution. We made use of the
excellent Lucene Information Retrieval SDK from the Apache
process, and a modified version of the Java
Matrix Package (JAMA) available from the NIST at Jama.
|
