<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#ffffff">
Hi Phaneendra,<br>
<br>
all points Sebastian wrote are valid. too much work in one project,
you can only do one think properly.<br>
<br>
Scientifically seen, you should focus on ONE hypothesis and test
that thouroughly.<br>
How we taught computer science over in Kaiserslautern our focus was
not about hacking as many things as sound cool and then describing
that they are so cool. Rather, hack one think and really dig into
proving it. (at least, that was the goal)<br>
<br>
Doing a recommender engine is always fruitful. <br>
<br>
As Sebastian says, you don't even have to store the explicit links.<br>
Its also easy to build a hypothesis.<br>
Its a CSE class...Evaluation? <br>
You may want to check out the evaluation work done within the PIM
community ("personal information management", they are on wikipedia
and have a mailinglist, I am a member. They do rock-solid
evaluations, serious researchers from MIT, Microsoft, ... good work)<br>
<br>
The search engine: check out this before you go on:<br>
<a class="moz-txt-link-freetext" href="http://www.springerlink.com/content/a6683407861281q2/">http://www.springerlink.com/content/a6683407861281q2/</a><br>
our (I am co-author there) idea was to NOT use NLP but just a
simplified proximity of words within the query. simple. works.<br>
still - the whole thing tooks years to hack and evaluate. don'T
underestimate NLP and querying, when done properly, it takes masses
of time.<br>
<br>
so Sebastian's comments remain: too many different ideas, go for one
and go for quality<br>
<br>
best<br>
Leo<br>
<br>
<br>
It was Sebastian Trüg who said at the right time 23.08.2011 16:07
the following words:
<blockquote cite="mid:4E53B408.8050307@kde.org" type="cite">
<pre wrap="">Hi Phaneendra,
some comments:
On 08/22/2011 10:39 PM, Phaneendra Hegde wrote:
</pre>
<blockquote type="cite">
<pre wrap="">*A Pure Semantic Desktop
*
The project has 2 major parts - Metadata processing and developing
applications using this powerful metadata.
*Processing Metadata of Nepomuk*
Right now, Linking of resources is not done automatically by
Nepomuk.We need to create tags for each resource manually. Since the
content of all files are already indexed and saved as part of the
Nepomuk metadata, we could use this to automatically tag all the resources.
Say, we have multiple files talking about 'technology', tags could be
created based on their content and link them to each other accordingly.
Similarly, resource linking could be carried out using other metadata
automatically. (Based on rating, file mime type, creation date and time
etc The nepomuk developer Vishesh Handa told me that he's also working
on automatic tagging of resources using NLP. But we try to extend this
idea by making use of these tagged resources. Tags can be used for
linking different resources automatically.
</pre>
</blockquote>
<pre wrap="">
Let me state that nao:Tag is intended for manual creation by the user.
Thus, if you want to do some auto-extraction of information rather use
something like pimo:Topic or even better specific types like
pimo:Project, pimo:Person, pimo:Location, pimo:Event, and so on.
In any case this sort of automatic relation creation is only of interest
if it uses actual NLP features which go beyond extracting keywords. By
that I mean disambiguation and topic detection without actually matching
the exact words.
Otherwise finding related resources on-the-fly would be much more
powerful without the problem of storing potentially wrong relations.
A query for the topics in the current resource could simply be done by
full-text search.
</pre>
<blockquote type="cite">
<pre wrap=""> Finally, this might lead to a more semantic desktop, where all
resources are inter related.A lot of new functionality can be added
using this information.
</pre>
</blockquote>
<pre wrap="">
Like what? What is the advantage over the on-the-fly approach I mention
above (given that you mostly do keyword-extraction)?
</pre>
<blockquote type="cite">
<pre wrap="">*Voice/text based desktop search engine
*The user asks ( or types in ) a question, and he is provided with all
the files/resources which match the criteria.
eg. He may ask 'Which movie did I watch last Saturday'
He will be provided with the list of movies he watched that day.
Natural Language processing will be used to process the question,
and using metadata stored in Virtuso, the results are generated.A common
voice to text converter could be used for audio queries.
People with physical disabilities could be benefited like this. It
would make the desktop usage more interactive adding a real life
perspective to
desktop usage. KDE doesn't have any voice based search as of now
(AFAIK), and this feature might become one of the key feature in future.
For searching , we have thought of providing "search and launch" kind of
plasma plugin/tool which can be used as searching tools like
synapse,gnome-do etc. More or less it might look like Google
search box , which provides voice based search with small voice input
icon on
right corner.
</pre>
</blockquote>
<pre wrap="">
I am not convinced by this. 1. This is a very hard topic and too much
for your rather short project alone, and 2. the hard part is extracting
the question from the text which requires a lot of powerful NLP. IMHO it
is not worth the effort at the moment.
</pre>
<blockquote type="cite">
<pre wrap="">*Intelligent file browser
</pre>
</blockquote>
<pre wrap="">
Also? IMHO this alone would be sufficient for the time you have.
</pre>
<blockquote type="cite">
<pre wrap="">*Plugin for file browser which suggests files/resources which are
similar to the file currently being used/selected.*
*
eg: Say user selects a song of a particular band, then he would be
suggested songs from the same band or any other metadata.
</pre>
</blockquote>
<pre wrap="">
A plugin system for this already exists in nepomukannotation and could
serve as a basis.
All in all I think you take on way too much. You should rather try to
trim it down and define in more detail what you want to accomplish.
Cheers,
Sebastian
</pre>
<blockquote type="cite">
<pre wrap="">This idea is the combination of 3-4 ideas which can be found at nepomuk
idea page <a class="moz-txt-link-rfc2396E" href="http://techbase.kde.org/Projects/Nepomuk"><http://techbase.kde.org/Projects/Nepomuk></a> .
I highly appreciate your comments and I'll be very happy to take your
suggestions regarding this project.
Thank you :)
Cheers
- PNH
Phaneendra Hegde
IRC- pnh @ freenode (#nepomuk-kde,#kde-in)
Blog: <a class="moz-txt-link-freetext" href="http://pnhblog.tumblr.com">http://pnhblog.tumblr.com</a>
</pre>
</blockquote>
<pre wrap="">_______________________________________________
Nepomuk mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Nepomuk@kde.org">Nepomuk@kde.org</a>
<a class="moz-txt-link-freetext" href="https://mail.kde.org/mailman/listinfo/nepomuk">https://mail.kde.org/mailman/listinfo/nepomuk</a>
</pre>
</blockquote>
<br>
<pre class="moz-signature" cols="72">--
Leo Sauermann, Dr.
CEO and Founder
mail: <a class="moz-txt-link-abbreviated" href="mailto:leo.sauermann@gnowsis.com">leo.sauermann@gnowsis.com</a>
mobile: +43 6991 gnowsis
Where your things come together,
Join: <a class="moz-txt-link-freetext" href="http://www.gnowsis.com/about/content/newsletter">http://www.gnowsis.com/about/content/newsletter</a>
Follow: <a class="moz-txt-link-freetext" href="http://twitter.com/Refinder">http://twitter.com/Refinder</a>
Like: <a class="moz-txt-link-freetext" href="http://www.facebook.com/Refinder">http://www.facebook.com/Refinder</a>
Learn: <a class="moz-txt-link-freetext" href="http://www.gnowsis.com/about/blog">http://www.gnowsis.com/about/blog</a>
Try: <a class="moz-txt-link-freetext" href="http://www.getrefinder.com/accounts/register/">http://www.getrefinder.com/accounts/register/</a>
____________________________________________________
</pre>
</body>
</html>