/Relevance-Model

Relevance based Language Model (Victor Lavrenko - SIGIR-2001) with query mix (RM3 - Jaleel et al. TREC-2004)

Primary LanguageJava

Relevance Based Language Model with Query mixing (RM3)

========

Build using Ant:

ant

Two ways to execute the program:

1. Using rblm.sh

<<<<<<< HEAD Set the following in rblm.sh file:

Set the following parameters in rblm.sh file:

4da92129f11520e842f2fb15c706ba6b2fc7141b

stopFilePath="The path of the stopword file"
indexPath="The path of the Lucene index on which the retrieval will be performed"
queryPath="The path of the query file in complete XML format"
resPath="Directory path in which the .res file will be saved in TREC 6-column .res format"

<<<<<<< HEAD

The similarity function that will be used for the actual retrieval is set to LMJelinekMercerSimilarity with parameter set to 0.6. (Modifiable)

numHits i.e. the maximum number of documents to retrieve is set to 1000. (Modifiable)

4da92129f11520e842f2fb15c706ba6b2fc7141b The shell script rblm.sh takes the following command line inputs:

1. Number of expansion documents
2. Number of expansion terms
<<<<<<< HEAD
3. RM3 QueryMix ([img]http://bit.ly/18aJOnT[/img]) for calculating [img]http://www.sciweavers.org/tex2img.php?eq=%5Clambda%20P%28w%7CR%29%20%2B%20%281-%5Clambda%29P%28w%7CQ%29&bc=White&fc=Black&im=jpg&fs=12&ff=arev&edit=0[/img]

Run:

./rblm.sh
=======
3. RM3 QueryMix LAMBDA for calculating LAMBDA*P(w|R) + (1-LAMBDA)*P(w|Q)
4. [Rerank] YES-1, NO-0; Default is 0.

Run:

./rblm.sh 20 70 0.6

The above parameter will apply the number of feedback documents and terms to 20 and 70 respectively and set the query-mix parameter to 0.6. Finally, there will be a re-retrieval with query expansion.

2. Directly running the RelevanceModel class

Create a properties file (say, init.properties) and set the following parameters:

indexPath=<The path of the Lucene index on which the retrieval will be performed>

queryPath=<The path of the query file in complete XML format>

stopFilePath=<The path of the stopword file>

resPath=<Directory path in which the .res file will be saved in TREC 6-column .res format>

numHits=<The maximum number of documents to retrieve>

### Similarity functions:
similarityFunction=0 OR 1 OR 2 OR 3 # Respectively for DefaultSimilarity, BM25Similarity, LMJelinekMercerSimilarity and LMDirichletSimilarity

param1=<First parameter of the similarity function>
param2=<Second parameter of the similarity function>

numFeedbackDocs=<Number of feedback documents>

numFeedbackTerms=<Number of feedback terms>

rm3.queryMix=<RM3 QueryMix `LAMBDA` for calculating `LAMBDA*P(w|R) + (1-LAMBDA)*P(w|Q)>`

rm3.rerank=true

Run:

java -Xmx6g -cp dist/RelevanceFeedback.jar RelevanceFeedback.RelevanceBasedLanguageModel init.properties
>>>>>>> 4da92129f11520e842f2fb15c706ba6b2fc7141b