Conference heron::euro_swas_ai

262.0. "Reprint: ITIR" by MYBABY::TARANTOLA (Carlo *AI IST* @VBE) Thu Jan 03 1991 00:31

            <<< SCAACT::APP$DISK:[NOTES$LIBRARY]EXCALIBUR.NOTE;1 >>>
                            -< Let's sell PixTex! >-
================================================================================
Note 46.0          Intelligent Text and Image Retrieval system        No replies
MYBABY::TARANTOLA "Carlo *AI IST* @VBE"             109 lines   2-JAN-1991 17:29
--------------------------------------------------------------------------------
    I put below the READ_ME file which comes with the kit. I hope it might
    answer some questions on what ITIR is about.
    
    	-Carlo
    
    --------
    
		Intelligent Text and Image Retrieval system (ITIR)

ITIR is a demonstration system that uses technology licensed to DIGITAL by 
Escalibur Technologies, Inc., McClean, VA, USA.

Excalibur Technologies Inc. licenses also two application products that 
customers can buy: they are TRS � (Text Retrival System) and PIXTEXT �. Both
are Digital Distributed Products.

Digital can offer product customizations based on a callable library. The ITIR
demostration is an example of such customization.

ITIR is written to demonstrate Excalibur technology in intelligent text and
image retrieval and it is intented for

			DIGITAL INTERNAL USE ONLY

Because of licensing issues, in order either to get a copy or to obtain more
information, you are invited to send mail to:

	Carlo YIPPEE (51.193)::TARANTOLA (Europe - @VBE) DTN: 828-5463 
or 	Ken   BUFFER (56.118)::HARRIS    (US     - @OGO) DTN: 276-9901

------------------------------------------------------------------------------

				User's Notes

ITIR allows the user to learn several ASCII file into an index file (.PIN).

ASCII files may be created using an editor, or by performing OCR on the images.

Every ASCII file can have one associated IMAGE file. A limitation is that you 
can not have more than one image associated to the same ASCII file. 
This image has to:
	- be in DDIF format
	- have the same name of the associated ASCII file (with a .DDIF 
	  extension for example)
	- be in the same directory of the associated ASCII file.

You are not required to move or copy your ASCII files neither before nor 
after ITIR learned them.
Of course if you modify your ASCII file you have to learn it again. At the
moment, in ITIR, you can not forget a file. This feature might be added in the
future if there is a sufficent request and I have time to do it.

To instal ITIR, do a 

	$ BACKUP saveset/sav []

in the directory you decided to use for ITIR and then executed the included
install command file. This file will create a ITIR$STARTUP.COM file in your
SYS$LOGIN directory. You need to run this file all the times you are logging 
in in order to define all the logicals and the symbol ITIR requires.

To use ITIR a normal way of doing things is:

	- Create a PIN file (if you never did it)
	- Learn as many ASCII file as you like
	- Retrieve information on the basis of clues you can enter from the
	  SEARCH menu
	- Save this information in a text file if you are happy with the search

In the PARAMETERS menu you have 3 submenus:

In the SEARCH PARAMS submenu:

You can set the maximum number of hits you want to display by changing the
"number of hits to display" parameter.

You can select the miminum number of characters that have to match your clue in
order to be highlighted, by modifying the "highligh sensitivity" parameter.

You can limit the use of high frequency patterns in scoring closeness of a 
search string to the pattern in the index. This is determined by the 
"max high frequency" parameter.

In the LEARN PARAMS submenu:

You can select the size of the sections into which an ASCII file is broken.
The bigger the section size the faster the answer but the worse the quality 
of the retrieved information.
The smaller the section size the slower the asnwer but the better the quality.

In the FILE PARAMS submenu:

These parameters are ignored at the moment.

In the SEARCH menu you have 3 submenus:

From the SEARCH submenu you can popup a window to enter your clue (a clue can
consist of more than one word. All the word are ORed together. In ITIR you do
not have any Boolean operators, but it is possible to introduce them if
needed).

From the RATE and NEURAL RATE submenus you can run a rating proces to obtain
a more precise score. It is basically a way to emphasize phrases. The
difference between the two is subtle and the best way to determine which works
best for you is to try both. However you have to be aware that NEURAL RATE is
slower than RATE.

	-Carlo