[Kim-discussion] if a named entity is an instance of two classes?

[Philip Alexiev]

Hi Mehnaz,

My opinion is that because of the great ambiguity of short literals and 
abbreviations the results will be quite noisy. You could try and do some 
experimenting. For that purpose you can use a standard Gate gazetteer 
and make a list with the abbreviations you want to recognize.

One thing that could increase the precision of the IE process is to take 
into consideration the context of the abbreviations. This is best 
achieved by the use of pattern matching of annotations and Jape grammars.

Greetings,
Philip



On 08/27/2009 04:14 AM, Mehnaz Adnan wrote:
> Hi Philip,
>
>
> I am trying to provide semantic annotation for the medical 
> abbreviations in my text. For this case I want to use only short forms 
> (eg. K) in IE process and provide long forms( e.g potassium)  in popup 
> window. For many cases I have more than one long forms. As I am 
> dealing with abbreviation so I  have a lot of phrases having less than 
> three symbols.
> Please advice me a good solution.
>
> Mehnaz
>
>
> 2009/8/26 Philip Alexiev <philip.alexiev at sirma.bg 
> <mailto:philip.alexiev at sirma.bg>>
>
>     Hi Mehnaz,
>
>     In our previous experience we have found that phrases which are
>     shorter than 3 symbols are too ambiguous and carry little benefit
>     to the IE process. That is why the gazetteer skips short labels on
>     initialization. It is true that in some special cases they might
>     be useful. That is why we have included a tool to customize the
>     init process of the gazetteer. It is present in version 3 of the
>     KIM platform.
>
>     If you provide some more info on the use case we will try to
>     provide a good solution.
>
>     Greetings,
>     Philip
>
>
>     On 08/26/2009 06:47 AM, Mehnaz Adnan wrote:
>>     Hi Philip
>>     I tried to use rdfs:comment property in this way
>>     <http://www.ontotext.com/kim/2006/05/wkb#K>
>>     <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
>>     <http://proton.semanticweb.org/2006/05/protont#MedicalAbbreviation> .
>>     <http://www.ontotext.com/kim/2006/05/wkb#K>
>>     <http://proton.semanticweb.org/2006/05/protons#generatedBy>
>>     <http://www.ontotext.com/kim/2006/05/wkb#Gazetteer> .
>>     <http://www.ontotext.com/kim/2006/05/wkb#K>
>>     <http://www.w3.org/2000/01/rdf-schema#label> "K" .
>>     <http://www.ontotext.com/kim/2006/05/wkb#K.001>
>>     <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
>>     <http://proton.semanticweb.org/2006/05/protons#Alias> .
>>     <http://www.ontotext.com/kim/2006/05/wkb#K.001>
>>     <http://www.w3.org/2000/01/rdf-schema#label> "K" .
>>     <http://www.ontotext.com/kim/2006/05/wkb#K>
>>     <http://proton.semanticweb.org/2006/05/protons#hasAlias>
>>     <http://www.ontotext.com/kim/2006/05/wkb#K.001> .
>>     <http://www.ontotext.com/kim/2006/05/wkb#K.001>
>>     <http://www.w3.org/2000/01/rdf-schema#comment> "Potassium" .
>>     But I does not generate annotation for the term "K" in my text,
>>     while it is generating annotations for all others in which I did
>>     not included comment property. I tried to do it by a custom
>>     property as well but i did not work.
>>     Please advice.
>>     Thanks and Regards
>>     Mehnaz
>>
>>     2009/8/24 Philip Alexiev <philip.alexiev at sirma.bg
>>     <mailto:philip.alexiev at sirma.bg>>
>>
>>         Hi Mehnaz,
>>
>>         It is conceptually incorrect to have instances of more than
>>         one class. My advice is to fix the model and not hack the
>>         recognition.
>>
>>         About the second question, it is strange to me that you don't
>>         want the complete name of the abbreviation recognized. A
>>         solution I would recommend, is to use the rdfs:comment
>>         property to describe the full alias. It will be shown as a
>>         description of the entity in the KB Explorer (popup window)
>>         and will not be used in the information extraction process.
>>         Another way is to use your custom property, which will be
>>         displayed in the list of properties for that entity.
>>
>>         Hope this helps,
>>         Philip
>>
>>
>>
>>         On 08/23/2009 09:46 PM, Mehnaz Adnan wrote:
>>>         Hi,
>>>         I have a named entities  "US" and "Ultrasound" both of these
>>>         are instances of two classes
>>>         <http://proton.semanticweb.org/2006/05/protont#InformationResource> 
>>>         and <http://proton.semanticweb.org/2006/05/protonu#Company>
>>>         When I view the annotations in GATE it annotates both "US"
>>>         and "Ultrasound" for "Organization" and not for "Object". I
>>>         am testing so just using KIM's ontologies for Populating
>>>         wkb.nt. I am attaching my wkb file as well.
>>>         Next question:
>>>         I want to use KIM to annotate abbreviations. For example AF
>>>         - Atrial Fibliration
>>>         for that I want to annotate "AF" and want to show "Atrial
>>>         Fibliration" as alias in KIM Web UI (Popup window). But I do
>>>         not want to annotate Atrial Fibliration  in the text. As I
>>>         understand KIM loaded labels from Aliases so I am Stuck how
>>>         to do it.
>>>         Please help
>>>
>>>         -- 
>>>         Mehnaz Adnan
>>>         Ph.D. Candidate,
>>>         Department of Computer Science-Tamaki
>>>         University of Auckland
>>>         email: madn002 at aucklanduni.ac.nz
>>>         <mailto:madn002 at aucklanduni.ac.nz>
>>
>>
>>
>>
>>
>>     -- 
>>     Mehnaz Adnan
>>     Ph.D. Candidate,
>>     Department of Computer Science-Tamaki
>>     University of Auckland
>>     Phone: 09-3737599 ext 83274
>>     email: madn002 at aucklanduni.ac.nz <mailto:madn002 at aucklanduni.ac.nz>
>
>
>
>
> -- 
> Mehnaz Adnan
> Ph.D. Candidate,
> Department of Computer Science-Tamaki
> University of Auckland
> Phone: 09-3737599 ext 83274
> email: madn002 at aucklanduni.ac.nz <mailto:madn002 at aucklanduni.ac.nz>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://ontotext.com/pipermail/kim-discussion/attachments/20090828/d409e88b/attachment-0002.html>