I am new to machine learning I am working to build a concept for Document Retrieval System using RANKLIB for boosting the positions of documents using click through data as Implicit feedback(count,time as features) by LAMBDAMART algorithm.
I have a querstion that do we need a ranking function to boost or is there any other alternative way to boost rank positions based on click through data.
Now the machine learns to change the positions as the training data will be updated whenever user makes new search.
Thanks in advance!
Last edit: sravan kumar karnati 2016-06-03
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi sravan kumar karnati ,
With RankLib you can build a model that reflects your training data.
As you were saying you can train a LambdaMARTModel .
At this point, you can re-rank and evaluate the model against test sets.
If you want to integrate this with your search engine, you need to implement the re-ranking using the model on your own.
Which technology are you using for the search engine ?
I have quite a large experience in integrating Learning To Rank in Solr.
Available to help,
Cheers
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi All,
I am also new to machine learning, and I am building a document retrieval system using the Lucene API, and I would link to use RankLib for document re-ranking.
Does RankLib has a Java API I could use? My search hasn't yielded anything so far.
Thanks!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
RankLib is more tuned toward adding different ranking algorithms to its own existing framework. There really isn't an API for using it in other applications.
You would have to write code that loads an existing, saved ranking model (from the available RankLib algorithms), and then essentially mirror existing RankLib classes and interfaces in your application to read in feature data (from a test file or generate your own features dynamically within your application), that then applies the saved model to rank results.
We are hoping to integrate RankLib saved models into the Galago search framework for ranking results that would likely involve creation of a sort of RankLib API, but that is not something that would be available soon.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello Everyone,
I am new to machine learning I am working to build a concept for Document Retrieval System using RANKLIB for boosting the positions of documents using click through data as Implicit feedback(count,time as features) by LAMBDAMART algorithm.
I have a querstion that do we need a ranking function to boost or is there any other alternative way to boost rank positions based on click through data.
Now the machine learns to change the positions as the training data will be updated whenever user makes new search.
Thanks in advance!
Last edit: sravan kumar karnati 2016-06-03
Hi sravan kumar karnati ,
With RankLib you can build a model that reflects your training data.
As you were saying you can train a LambdaMARTModel .
At this point, you can re-rank and evaluate the model against test sets.
If you want to integrate this with your search engine, you need to implement the re-ranking using the model on your own.
Which technology are you using for the search engine ?
I have quite a large experience in integrating Learning To Rank in Solr.
Available to help,
Cheers
Hi All,
I am also new to machine learning, and I am building a document retrieval system using the Lucene API, and I would link to use RankLib for document re-ranking.
Does RankLib has a Java API I could use? My search hasn't yielded anything so far.
Thanks!
RankLib is more tuned toward adding different ranking algorithms to its own existing framework. There really isn't an API for using it in other applications.
You would have to write code that loads an existing, saved ranking model (from the available RankLib algorithms), and then essentially mirror existing RankLib classes and interfaces in your application to read in feature data (from a test file or generate your own features dynamically within your application), that then applies the saved model to rank results.
We are hoping to integrate RankLib saved models into the Galago search framework for ranking results that would likely involve creation of a sort of RankLib API, but that is not something that would be available soon.
Hi Stephen,
Can you please provide more information on how I can achieve this?
Thank you
There are actually plugins to Solr (i.e., Lucene) that can do this, some of
which use RankLib. E.g.,
Hth,
Edgar
On Fri, Jul 1, 2016 at 8:13 PM Kwabena Mensa-Bonsu kid75@users.sf.net
wrote:
Thank you!