---

SimpleKLRetMethod Class Reference

KL Divergence retrieval model with simple document model smoothing. More...

#include <SimpleKLRetMethod.hpp>

Inheritance diagram for SimpleKLRetMethod:

List of all members.

Public Methods
	SimpleKLRetMethod (const Index &dbIndex, const string &supportFileName, ScoreAccumulator &accumulator)
	Construction of SimpleKLRetMethod requires a smoothing support file, which can be generated by the application GenerateSmoothSupport. The use of this smoothing support file is to store some pre-computed quantities so that the scoring procedure can be speeded up.
virtual	~SimpleKLRetMethod ()
virtual TextQueryRep *	computeTextQueryRep (const TermQuery &qry)
	compute the query representation for a text query (caller responsible for deleting the memory of the generated new instance)
virtual DocumentRep *	computeDocRep (DOCID_T docID)
	compute the doc representation (caller responsible for deleting the memory of the generated new instance)
virtual ScoreFunction *	scoreFunc ()
	return the scoring function pointer
virtual void	updateTextQuery (TextQueryRep &origRep, const DocIDSet &relDocs)
	Modify/update the query representation based on a set (presumably) relevant documents.
void	setDocSmoothParam (SimpleKLParameter::DocSmoothParam &docSmthParam)
void	setQueryModelParam (SimpleKLParameter::QueryModelParam &queryModParam)
Protected Methods
void	loadSupportFile ()
	Load support file support.
query model updating methods (i.e., feedback methods)
void	computeMixtureFBModel (SimpleKLQueryModel &origRep, const DocIDSet &relDocs)
	Mixture model feedback method.
void	computeDivMinFBModel (SimpleKLQueryModel &origRep, const DocIDSet &relDocs)
	Divergence minimization feedback method.
void	computeMarkovChainFBModel (SimpleKLQueryModel &origRep, const DocIDSet &relDocs)
	Markov chain feedback method.
void	computeRM1FBModel (SimpleKLQueryModel &origRep, const DocIDSet &relDocs)
	Relevance model1 feedback method.
void	computeRM2FBModel (SimpleKLQueryModel &origRep, const DocIDSet &relDocs)
	Relevance model1 feedback method.
Protected Attributes
double *	mcNorm
	needed for fast one-step Markov chain
double *	docProbMass
	needed for fast alpha computing
COUNT_T *	uniqueTermCount
	needed for supporting fast absolute discounting
UnigramLM *	collectLM
	a little faster if pre-computed
DocUnigramCounter *	collectLMCounter
	support the construction of collectLM
SimpleKLScoreFunc *	scFunc
	keep a copy to be used at any time
SimpleKLParameter::DocSmoothParam	docParam
SimpleKLParameter::QueryModelParam	qryParam
const string	supportFile

---

Detailed Description

KL Divergence retrieval model with simple document model smoothing.

---

Constructor & Destructor Documentation

SimpleKLRetMethod::SimpleKLRetMethod (  const Index &    dbIndex, const string &    supportFileName, ScoreAccumulator &    accumulator )

Construction of SimpleKLRetMethod requires a smoothing support file, which can be generated by the application GenerateSmoothSupport. The use of this smoothing support file is to store some pre-computed quantities so that the scoring procedure can be speeded up.

SimpleKLRetMethod::~SimpleKLRetMethod (    )  [virtual]

---

Member Function Documentation

void SimpleKLRetMethod::computeDivMinFBModel (  SimpleKLQueryModel &    origRep, const DocIDSet &    relDocs )  [protected]

Divergence minimization feedback method.

DocumentRep * SimpleKLRetMethod::computeDocRep (  DOCID_T    docID )  [virtual]

compute the doc representation (caller responsible for deleting the memory of the generated new instance) Implements TextQueryRetMethod.

void SimpleKLRetMethod::computeMarkovChainFBModel (  SimpleKLQueryModel &    origRep, const DocIDSet &    relDocs )  [protected]

Markov chain feedback method.

void SimpleKLRetMethod::computeMixtureFBModel (  SimpleKLQueryModel &    origRep, const DocIDSet &    relDocs )  [protected]

Mixture model feedback method.

void SimpleKLRetMethod::computeRM1FBModel (  SimpleKLQueryModel &    origRep, const DocIDSet &    relDocs )  [protected]

Relevance model1 feedback method.

void SimpleKLRetMethod::computeRM2FBModel (  SimpleKLQueryModel &    origRep, const DocIDSet &    relDocs )  [protected]

Relevance model1 feedback method.

virtual TextQueryRep* SimpleKLRetMethod::computeTextQueryRep (  const TermQuery &    qry )  [inline, virtual]

compute the query representation for a text query (caller responsible for deleting the memory of the generated new instance) Implements TextQueryRetMethod.

void SimpleKLRetMethod::loadSupportFile (    )  [protected]

Load support file support.

virtual ScoreFunction* SimpleKLRetMethod::scoreFunc (    )  [inline, virtual]

return the scoring function pointer Implements TextQueryRetMethod.

void SimpleKLRetMethod::setDocSmoothParam (  SimpleKLParameter::DocSmoothParam &    docSmthParam )  [inline]

void SimpleKLRetMethod::setQueryModelParam (  SimpleKLParameter::QueryModelParam &    queryModParam )  [inline]

void SimpleKLRetMethod::updateTextQuery (  TextQueryRep &    origRep, const DocIDSet &    relDocs )  [virtual]

Modify/update the query representation based on a set (presumably) relevant documents. Implements TextQueryRetMethod.

---

Member Data Documentation

UnigramLM* SimpleKLRetMethod::collectLM [protected]

a little faster if pre-computed

DocUnigramCounter* SimpleKLRetMethod::collectLMCounter [protected]

support the construction of collectLM

SimpleKLParameter::DocSmoothParam SimpleKLRetMethod::docParam [protected]

double* SimpleKLRetMethod::docProbMass [protected]

needed for fast alpha computing

double* SimpleKLRetMethod::mcNorm [protected]

needed for fast one-step Markov chain

SimpleKLParameter::QueryModelParam SimpleKLRetMethod::qryParam [protected]

SimpleKLScoreFunc* SimpleKLRetMethod::scFunc [protected]

keep a copy to be used at any time

const string SimpleKLRetMethod::supportFile [protected]

COUNT_T* SimpleKLRetMethod::uniqueTermCount [protected]

needed for supporting fast absolute discounting

---

The documentation for this class was generated from the following files:

• SimpleKLRetMethod.hpp
• SimpleKLRetMethod.cpp

---