edu.ucla.sspace.wordsi

Interface DependencyContextGenerator

All Known Implementing Classes:

AbstractOccurrenceDependencyContextGenerator, OccurrenceDependencyContextGenerator, OrderingDependencyContextGenerator, PartOfSpeechDependencyContextGenerator, RandomIndexingDependencyContextGenerator, SelPrefDependencyContextGenerator, WordOccrrenceDependencyContextGenerator

public interface DependencyContextGenerator

An interface for generating context vectors from raw unparsed text. A context vector can be generated from a sliding window of text. ContextGenerators are a subcomponent of ContextExtractors. As a ContextExtractor examines text and finds a word which is being represented in the Wordsi space, it will call the ContextGenerator and pass the created context vector to a Wordsi implementation.

ContextGenerators are recomended to be made serializable. They will serve as the core representation method of Wordsi implementations and can thus be re-used in multiple evaluations. For example, after training a Wordsi model, it may need to be evaluated in a psuedo-word disambiguation task or a SemEval task. In both cases, the feature space must remain the same turing training and evaluation.

For evaluation purposes, an added option is available: a read only mode. When in read only mode, ContextGenerators should not create any new features. If some co-occurring term does not exist in the feature space, it should be left out of the context vector, only feature which already exist in the space should contribute to the context vector. In standard mode, the generator is permitted to decided which words should serve as features using any method.

Author:

Keith Stevens

Method Summary

Methods

Modifier and Type	Method and Description
SparseDoubleVector	generateContext(DependencyTreeNode[] tree, int focusIndex) Returns a SparseDoubleVector that represents the context composed of the set of prevWords before the focus word and the set of nextWords after the focus word.
int	getVectorLength() Returns the maximum number of dimensions used to represent any given context.
void	setReadOnly(boolean readOnly) Sets the read only mode of the ContextGenerator.

Method Detail

generateContext

SparseDoubleVector generateContext(DependencyTreeNode[] tree,
                                 int focusIndex)

Returns a SparseDoubleVector that represents the context composed of the set of prevWords before the focus word and the set of nextWords after the focus word. Since sparse vectors are returned, if a second order vector is generated, it is recommended that the vector also be sparsed or have very few dimensions.

getVectorLength

int getVectorLength()

Returns the maximum number of dimensions used to represent any given context.

setReadOnly

void setReadOnly(boolean readOnly)

Sets the read only mode of the ContextGenerator. When set to read only, it prevents any new features from being generated.