Re: Introducing FRED

Hi all, and thanks Kingsley for starting this thread. 
FRED does many things actually, attempting to provide a decently deep machine reading (aka knowledge extraction from text) for the semantic web and linked data. 

Its pipeline includes components for several NLP tasks: named entity recognition and resolution, sense tagging and disambiguation, taxonomy induction, coreference resolution, semantic role labeling, event extraction, frame detection, negation and modality recognition, and a few more.

Of course, it relies on multiple, heterogeneous NLP techniques, and merges them by using some assumptions on how to map lexical and discourse semantics into semantic web ontology design.

It has been evaluated on some different tasks until now: frame detection [1], sense tagging, disambiguation, and taxonomy induction [2], see also the Těpalo tool at [4], and semantic sentiment analysis [3], see also the Sentilo tool at [5]. Evaluation of ontology learning (in the sense of pure schema learning) has not yet been performed.

In all cases, so many tasks inevitably lead to "error mixing": the main FRED component does not insert a significant amount of error, but individual peripheral components (specially sytactic parsing, named entity recognition, and word sense disambiguation) typically come with a certain amount of errors. In order to have an idea of accuracy, please refer to [6], which is a landscape analysis (not a full evaluation) of the performance of a dozen knowledge extraction tools when applied to semantic web tasks.
The specific "sameAs" error is probably due to a "tolerant" setting for Stanbol enhancers wrt named entity resolution, by no means to FRED core.

Also consider that peripheral components of FRED are not fixed, but they can be plugged based on the Apache Stanbol [7] RESTful interface (developed within the EU IKS project [8]), please write for details. The online version currently uses Stanbol enhancers for NER and IMS for WSD, but they will be made configurable in the future.

Ciao
Aldo

[1] http://stlab.istc.cnr.it/documents/papers/fred-ekaw12.pdf
[2] https://files.ifi.uzh.ch/ddis/iswc_archive/iswc/ab/2013/iswc2012.semanticweb.org/sites/default/files/76490065.pdf
[3] IEEE Computational Intelligence, to appear
[4] http://wit.istc.cnr.it/stlab-tools/tipalo
[5] http://wit.istc.cnr.it/stlab-tools/sentilo
[6] http://eswc-conferences.org/sites/default/files/papers2013/gangemi.pdf
[7] http://stanbol.apache.org/
[8] http://iks-project.eu

On Oct 31, 2013, at 6:54:04 AM , Kingsley Idehen <kidehen@openlinksw.com> wrote:

> On 10/30/13 3:30 PM, Krzysztof Janowicz wrote:
>> Hmm, the part where you are sameAs the class Person is very supridsing to me.
>> Krzysztof
> 
> I think the PNG that it generated created that illusion [1]. I scanned through the generated Turtle [2] and all I see is:
> 
> <http://www.ontologydesignpatterns.org/ont/fred/domain.owl#Kingsley_idehen>
>      a <http://www.ontologydesignpatterns.org/ont/fred/domain.owl#Person> ;
> <http://www.ontologydesignpatterns.org/ont/boxer/boxer.owl#possibleType>
>              <http://xmlns.com/foaf/0.1/Person> ;
> <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#hasQuality>
> <http://www.ontologydesignpatterns.org/ont/fred/domain.owl#Interested> , <http://www.ontologydesignpatterns.org/ont/fred/domain.owl#Male> ;
>      <http://www.w3.org/2002/07/owl#sameAs>
>              <http://dbpedia.org/resource/Person> , <http://dbpedia.org/resource/Kingsley_Uyi_Idehen> , <http://dbpedia.org/resource/Kingsley_Idehen> .
> 
> Links:
> 
> [1] http://bit.ly/1h171LT -- PNG
> [2] http://bit.ly/HsEP52 -- Turtle Doc .
> 
> Kingsley
>> 
>> On 10/30/2013 12:11 PM, Kingsley Idehen wrote:
>>> All,
>>> 
>>> I stumbled across FRED [1] via my daily tweet stream. I've had a quick play with this tool and I think most of you will find it quite interesting. Basically, it generates an OWL ontology from simple sentences [2][3][4].
>>> 
>>> Links:
>>> 
>>> [1] http://wit.istc.cnr.it/stlab-tools/fred -- project home page
>>> [2] http://t.co/edHh2N0N9e -- simple sentences describing some of my likes and dislikes
>>> [3] http://t.co/0AqMmBI2BA -- PNG visualization of the generated ontology
>>> [4] http://bit.ly/HsEP52 -- Turtle representation of the generated ontology .
>>> 
>> 
>> 
> 
> 
> -- 
> 
> Regards,
> 
> Kingsley Idehen	
> Founder & CEO
> OpenLink Software
> Company Web: http://www.openlinksw.com
> Personal Weblog: http://www.openlinksw.com/blog/~kidehen
> Twitter/Identi.ca handle: @kidehen
> Google+ Profile: https://plus.google.com/112399767740508618350/about
> LinkedIn Profile: http://www.linkedin.com/in/kidehen
> 
> 
> 
> 
>

Received on Thursday, 31 October 2013 00:29:20 UTC