You are right. I saw your email after pressing send. Let me
experiment. Thanks for the tip.
Best,
Amit
On Jul 12, 2006, at 10:55 AM, mark harwood wrote:
Appending POS to the terms will create post processing nightmare
I think you may have missed the subtle distinction between Grant's
suggestion and mine.
His suggestion was to append your POS info to the source token -
creating a single token which combined both the original content
and your POS info.
My suggestion was to generate *two* tokens - one is the original
source token, untouched, the other is the POS token but the trick
is to set the "position increment" to 0 for the POS token. This
ensures that the POS token is considered to occur at exactly the
same location as the source token. The original source token is
untampered with and has all the usual stats in the index.The POS
token and/or source token can be used interchangeably in queries on
the same field so you can use position info eg phrase or span queries.
Cheers
Mark
----- Original Message ----
From: Amit Kumar <[EMAIL PROTECTED]>
To: java-user@lucene.apache.org
Sent: Wednesday, 12 July, 2006 4:15:34 PM
Subject: Re: Storing Part of Speech information in Lucene Indices
We need to be able to search by word and POS and also have POS
available for each occurrence. Appending POS to the terms will
create post processing nightmare to retrieve
term frequencies right? (I would have to add all the foo_NN and
foo_ADJ etc.).
I can store the POS in a parallel field and access it via term
vectors, but that wouldn't allow any kind of search on POS related
fields right? For example if I wanted to search for any
adjective with in 3 words of say a term or say If I wanted to get all
the patterns that follow the sequence ADJ NN ADJ.
Let me look in the developer archives for the payload discussions,
perhaps implementing that might satisfy my use cases.
Comments?
-Thanks
Amit
On Jul 12, 2006, at 6:39 AM, Grant Ingersoll wrote:
Hi Amit,
This is definitely something you can do. What are your goals for
it? Do you want to search by word and POS or do you just want POS
available for post processing?
You could just append the POS tag onto the end of your token as it
gets indexed, something like foo_NN or foo_ADJ. This approach may
mean you have to use prefix query when you want to search against
just "foo". You could also have a parallel field to your main
field that stores the POS. Then you could access it via the term
vectors array.
Also, we have been discussing on the developers list on how to add
payloads to a posting (i.e. store related information at a position
in the index) similar to what Google discusses in their original
paper. Unfortunately, this isn't implemented yet, but if you feel
like helping out, check out the discussion on the developer's list
(see Flexible Indexing).
-Grant
On Jul 12, 2006, at 1:36 AM, Amit Kumar wrote:
Hi,
A new project that I am investigating lucene for needs the Parts
of speech information for the tokens. I can get that
information using NLP techniques (GATE etc.), by pre processing
the documents but I would like to store that
information in the Indices. Something along the lines of
TermVectorOffsetInfo[?].getPartofSpeech();
I am writing to ask for your advice, you can tell me I am b o n k
e r s or let me know where I should start digging :).
Is that a good idea? Or would it be just less trouble for me to
store the offset information along with parts of speech
outside Lucene.
Has anyone else done that?
Best,
Amit
ps: Thank you for putting the LuceneInAction source online, it was
a great help to see the CategorizerTest.java.
I am ordering my copy of the book tomorrow :)
---------------------------------------------------------
Amit Kumar
Research Programmer
The Graduate School of Library and Information Science
University of Illinois, Urbana Champaign IL, 61820
phone: 217-333-4118 fax: 217-244-3302
---------------------------------------------------------
--------------------------
Grant Ingersoll
Sr. Software Engineer
Center for Natural Language Processing
Syracuse University
335 Hinds Hall
Syracuse, NY 13244
http://www.cnlp.org
Voice: 315-443-5484
Fax: 315-443-6886
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------
Amit Kumar
Research Programmer
The Graduate School of Library and Information Science
University of Illinois, Urbana Champaign IL, 61820
phone: 217-333-4118 fax: 217-244-3302
---------------------------------------------------------
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------
Amit Kumar
Research Programmer
The Graduate School of Library and Information Science
University of Illinois, Urbana Champaign IL, 61820
phone: 217-333-4118 fax: 217-244-3302
---------------------------------------------------------