You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@stanbol.apache.org by Tharindu Rusira <th...@gmail.com> on 2014/03/04 13:20:01 UTC
Topic classification engine for Stanbol
Hi all,
As mentioned in STANBOL-197[1], a topic classification engine for Stanbol
is being developed. In [2], it is discussed that this engine uses Solr
based classifier and the possibility of introducing more rigorous
classifiers based on OpenNLP and Mahout.
I am interested in this idea and like to contribute for future
developments. So I would like to know the current status of the project.
[1] https://issues.apache.org/jira/browse/STANBOL-197
[2] https://vimeo.com/45633053
Thanks,
--
M.P. Tharindu Rusira Kumara
Department of Computer Science and Engineering,
University of Moratuwa,
Sri Lanka.
+94757033733
www.tharindu-rusira.blogspot.com
Re: Topic classification engine for Stanbol
Posted by Rafa Haro <rh...@apache.org>.
Hi Tharindu, Rupert
El 12/03/14 09:07, Rupert Westenthaler escribió:
> I like the general Architecture containing of a
>
> * TopicClassifier
> * TrainingSet
>
> as this allows to have different implementations of managing the
> training set (e.g. in Solr, a RDF tripleStore, a database or simple
> files in a file system) and TopicClassifiers (Solr, OpenNLP, Mahout,
> ...) Note also the the trainingSet part is optional and only required
> for TopicClassifier that can dynamically update their classification
> models.
Regarding this, maybe it is important again to bring to the scene a new
version of the CMS Adapter and ContentHub which could easily feed the
training API.
Cheers, Rafa
Re: Topic classification engine for Stanbol
Posted by Rupert Westenthaler <ru...@gmail.com>.
Hi Tharindu,
Totally forgot about STANBOL-1278. Fixed it today in the morning.
Thanks for the remainder.
Related to the architecture:
I like the general Architecture containing of a
* TopicClassifier
* TrainingSet
as this allows to have different implementations of managing the
training set (e.g. in Solr, a RDF tripleStore, a database or simple
files in a file system) and TopicClassifiers (Solr, OpenNLP, Mahout,
...) Note also the the trainingSet part is optional and only required
for TopicClassifier that can dynamically update their classification
models.
The interfaces itself will need to be adapted/improved as more
implementations will be added. The current API looks a bit tailored to
the Solr based implementation.
If possible I would like to have Cross Validation to be implemented in
an implementation independent way.
please also note STANBOL-1294 [1]
best
Rupert
[1] https://issues.apache.org/jira/browse/STANBOL-1294
On Tue, Mar 11, 2014 at 2:04 PM, Tharindu Rusira
<th...@gmail.com> wrote:
> On Tue, Mar 4, 2014 at 8:06 PM, Rupert Westenthaler <
> rupert.westenthaler@gmail.com> wrote:
>
>> Hi Tharindu
>>
>> Thanks for you interest!
>>
>> > As mentioned in STANBOL-197[1], a topic classification engine for Stanbol
>> > is being developed. In [2], it is discussed that this engine uses Solr
>> > based classifier and the possibility of introducing more rigorous
>> > classifiers based on OpenNLP and Mahout.
>> > I am interested in this idea and like to contribute for future
>> > developments. So I would like to know the current status of the project.
>>
>> The most recent release of the Topic Classification Engine mentioned
>> in [1] and [2] contains of thee modules [3].
>>
>> NOTE that the trunk version of this engine (1.0.0-SNAPSHOT) has still
>> an open issue [4] that breaks the RESTful API used for training.
>>
>> Hi Rupert,
> Thanks for your suggestions and I would like to clarify further questions
> regarding the existing topic classifier.
>
> The said issue(STANBOL-1278) occurs at the run time and apparently it is
> due to an osgi bundle dependency resolution issue. So what I was thinking
> is, if we are integrating tools as I've suggested in my previous mail, do
> we have to adhere to the existing architecture or is it desirable to go for
> a new topic classification design?
> Why I'm asking this is, if we decide to stick to the existing architecture,
> resolving STANBOL-1278 will also be a part of the required workflow.
> Further, the current design was done to integrate a Solr instance. So we
> need to consider the possibility/feasibility of using the existing design
> to integrate Mahout and OpenNLP.
>
> Your opinion on this matter is highly appreciated.
>
> Thanks,
>
>
>
>> AFAIK no work regarding OpenNLP or Mahout was done. If you want to
>> implement other topic classification engine you can start from the
>> Enhancement-Engine archetype [5]. Topic engines are expected to
>> contribute fise:TopicAnnotation [6] to the metadata of the content
>> item.
>>
>> Having other topic classification engine would be really great. If you
>> have questions or need some help feel free to ask here or directly in
>> the #stanbol channel on freenode.
>>
>> best
>> Rupert
>>
>>
>> > [1] https://issues.apache.org/jira/browse/STANBOL-197
>> > [2] https://vimeo.com/45633053
>> [3] http://search.maven.org/#search|ga|1|stanbol%20topic
>> [4] https://issues.apache.org/jira/browse/STANBOL-1278
>> [5]
>> http://svn.apache.org/repos/asf/stanbol/trunk/development/archetypes/enhancement-engine/
>> [6]
>> http://stanbol.apache.org/docs/trunk/components/enhancer/enhancementstructure.html#fisetopicannotation
>>
>> On Tue, Mar 4, 2014 at 1:20 PM, Tharindu Rusira
>> <th...@gmail.com> wrote:
>> > Hi all,
>> > As mentioned in STANBOL-197[1], a topic classification engine for Stanbol
>> > is being developed. In [2], it is discussed that this engine uses Solr
>> > based classifier and the possibility of introducing more rigorous
>> > classifiers based on OpenNLP and Mahout.
>> > I am interested in this idea and like to contribute for future
>> > developments. So I would like to know the current status of the project.
>> >
>> > [1] https://issues.apache.org/jira/browse/STANBOL-197
>> > [2] https://vimeo.com/45633053
>> >
>> > Thanks,
>> >
>> > --
>> > M.P. Tharindu Rusira Kumara
>> >
>> > Department of Computer Science and Engineering,
>> > University of Moratuwa,
>> > Sri Lanka.
>> > +94757033733
>> > www.tharindu-rusira.blogspot.com
>>
>>
>>
>> --
>> | Rupert Westenthaler rupert.westenthaler@gmail.com
>> | Bodenlehenstraße 11 ++43-699-11108907
>> | A-5500 Bischofshofen
>>
>
>
>
> --
> M.P. Tharindu Rusira Kumara
>
> Department of Computer Science and Engineering,
> University of Moratuwa,
> Sri Lanka.
> +94757033733
> www.tharindu-rusira.blogspot.com
--
| Rupert Westenthaler rupert.westenthaler@gmail.com
| Bodenlehenstraße 11 ++43-699-11108907
| A-5500 Bischofshofen
Re: Topic classification engine for Stanbol
Posted by Tharindu Rusira <th...@gmail.com>.
On Tue, Mar 4, 2014 at 8:06 PM, Rupert Westenthaler <
rupert.westenthaler@gmail.com> wrote:
> Hi Tharindu
>
> Thanks for you interest!
>
> > As mentioned in STANBOL-197[1], a topic classification engine for Stanbol
> > is being developed. In [2], it is discussed that this engine uses Solr
> > based classifier and the possibility of introducing more rigorous
> > classifiers based on OpenNLP and Mahout.
> > I am interested in this idea and like to contribute for future
> > developments. So I would like to know the current status of the project.
>
> The most recent release of the Topic Classification Engine mentioned
> in [1] and [2] contains of thee modules [3].
>
> NOTE that the trunk version of this engine (1.0.0-SNAPSHOT) has still
> an open issue [4] that breaks the RESTful API used for training.
>
> Hi Rupert,
Thanks for your suggestions and I would like to clarify further questions
regarding the existing topic classifier.
The said issue(STANBOL-1278) occurs at the run time and apparently it is
due to an osgi bundle dependency resolution issue. So what I was thinking
is, if we are integrating tools as I've suggested in my previous mail, do
we have to adhere to the existing architecture or is it desirable to go for
a new topic classification design?
Why I'm asking this is, if we decide to stick to the existing architecture,
resolving STANBOL-1278 will also be a part of the required workflow.
Further, the current design was done to integrate a Solr instance. So we
need to consider the possibility/feasibility of using the existing design
to integrate Mahout and OpenNLP.
Your opinion on this matter is highly appreciated.
Thanks,
> AFAIK no work regarding OpenNLP or Mahout was done. If you want to
> implement other topic classification engine you can start from the
> Enhancement-Engine archetype [5]. Topic engines are expected to
> contribute fise:TopicAnnotation [6] to the metadata of the content
> item.
>
> Having other topic classification engine would be really great. If you
> have questions or need some help feel free to ask here or directly in
> the #stanbol channel on freenode.
>
> best
> Rupert
>
>
> > [1] https://issues.apache.org/jira/browse/STANBOL-197
> > [2] https://vimeo.com/45633053
> [3] http://search.maven.org/#search|ga|1|stanbol%20topic
> [4] https://issues.apache.org/jira/browse/STANBOL-1278
> [5]
> http://svn.apache.org/repos/asf/stanbol/trunk/development/archetypes/enhancement-engine/
> [6]
> http://stanbol.apache.org/docs/trunk/components/enhancer/enhancementstructure.html#fisetopicannotation
>
> On Tue, Mar 4, 2014 at 1:20 PM, Tharindu Rusira
> <th...@gmail.com> wrote:
> > Hi all,
> > As mentioned in STANBOL-197[1], a topic classification engine for Stanbol
> > is being developed. In [2], it is discussed that this engine uses Solr
> > based classifier and the possibility of introducing more rigorous
> > classifiers based on OpenNLP and Mahout.
> > I am interested in this idea and like to contribute for future
> > developments. So I would like to know the current status of the project.
> >
> > [1] https://issues.apache.org/jira/browse/STANBOL-197
> > [2] https://vimeo.com/45633053
> >
> > Thanks,
> >
> > --
> > M.P. Tharindu Rusira Kumara
> >
> > Department of Computer Science and Engineering,
> > University of Moratuwa,
> > Sri Lanka.
> > +94757033733
> > www.tharindu-rusira.blogspot.com
>
>
>
> --
> | Rupert Westenthaler rupert.westenthaler@gmail.com
> | Bodenlehenstraße 11 ++43-699-11108907
> | A-5500 Bischofshofen
>
--
M.P. Tharindu Rusira Kumara
Department of Computer Science and Engineering,
University of Moratuwa,
Sri Lanka.
+94757033733
www.tharindu-rusira.blogspot.com
Re: Topic classification engine for Stanbol
Posted by Rupert Westenthaler <ru...@gmail.com>.
Hi Tharindu
Thanks for you interest!
> As mentioned in STANBOL-197[1], a topic classification engine for Stanbol
> is being developed. In [2], it is discussed that this engine uses Solr
> based classifier and the possibility of introducing more rigorous
> classifiers based on OpenNLP and Mahout.
> I am interested in this idea and like to contribute for future
> developments. So I would like to know the current status of the project.
The most recent release of the Topic Classification Engine mentioned
in [1] and [2] contains of thee modules [3].
NOTE that the trunk version of this engine (1.0.0-SNAPSHOT) has still
an open issue [4] that breaks the RESTful API used for training.
AFAIK no work regarding OpenNLP or Mahout was done. If you want to
implement other topic classification engine you can start from the
Enhancement-Engine archetype [5]. Topic engines are expected to
contribute fise:TopicAnnotation [6] to the metadata of the content
item.
Having other topic classification engine would be really great. If you
have questions or need some help feel free to ask here or directly in
the #stanbol channel on freenode.
best
Rupert
> [1] https://issues.apache.org/jira/browse/STANBOL-197
> [2] https://vimeo.com/45633053
[3] http://search.maven.org/#search|ga|1|stanbol%20topic
[4] https://issues.apache.org/jira/browse/STANBOL-1278
[5] http://svn.apache.org/repos/asf/stanbol/trunk/development/archetypes/enhancement-engine/
[6] http://stanbol.apache.org/docs/trunk/components/enhancer/enhancementstructure.html#fisetopicannotation
On Tue, Mar 4, 2014 at 1:20 PM, Tharindu Rusira
<th...@gmail.com> wrote:
> Hi all,
> As mentioned in STANBOL-197[1], a topic classification engine for Stanbol
> is being developed. In [2], it is discussed that this engine uses Solr
> based classifier and the possibility of introducing more rigorous
> classifiers based on OpenNLP and Mahout.
> I am interested in this idea and like to contribute for future
> developments. So I would like to know the current status of the project.
>
> [1] https://issues.apache.org/jira/browse/STANBOL-197
> [2] https://vimeo.com/45633053
>
> Thanks,
>
> --
> M.P. Tharindu Rusira Kumara
>
> Department of Computer Science and Engineering,
> University of Moratuwa,
> Sri Lanka.
> +94757033733
> www.tharindu-rusira.blogspot.com
--
| Rupert Westenthaler rupert.westenthaler@gmail.com
| Bodenlehenstraße 11 ++43-699-11108907
| A-5500 Bischofshofen