The Scientific World Journal / 2015 / Article / Tab 1

Review Article

Lightweight Adaptation of Classifiers to Users and Contexts: Trends of the Emerging Domain

Table 1

Characteristics of the reviewed adaptation approaches. Notations: X: yes; A: depending on a chosen algorithm; for example, training data can be used instead of assumptions about new context and unlabelled data are used only as negative examples; MK: multiclass classification problems only; U: data are vectors of user preferences; base classifier: either ensemble member or the only context-specific model.

Method nameApplicabilityTrainingAdditional data
Suits dissimilar contextsSuits large varieties of problems/algorithmsMost lightweightRequires little assumptions about new contextsFor training of base classifier(s)For classifier selection/combinationUses contextual dataUses raw primary data of other contextsUses models for other contextsUses unlabelled data for the target context

Model selection
Contextual weightingXXAAXXA
Optimising utility functionXAAAA
Tuning classifiers for small datasetsXXXA
Cascaded trainingXXXXX
Learning context-specific relations between classifier outputsXMKXXXA
Optimising model parameters with evolutionary algorithmsXXXXXAX
Optimising model parameters with gradient descentAXAXAX
Algorithm-specific methods to shift a decision boundaryAXXAXA
Adapting only selected parametersXXXXXAX
Error weightingXXXXA
The use of model parameters as training dataAXX
Vector modificationXUXXXXA
Modifying a similarity measureXUAAXAAA
Target context-specific combinations of cues, obtained in other contextsXUAXXAAA

Ensembles
Factor ensemblesXXXXA
Diversity-based ensemblesXXXXA
Optimising a pool of classifiers, trained on data for several contextsXXXXX
Ensemble of generalisersXXXX
Knowledge transfer ensemblesXXXXXAXAA
Stacked ensemblesXXXX
Dynamic selection of base classifiersXXAXXAA
Sample-selecting ensemblesXXAXXAA

Context as a feature
Embedding contextual parameters as additional nodes into graphical modelsXXXXA
Using historical contexts as nodes in graphical modelsXXAXA
Using contextual parameters as input featuresXXXXXXXA
Including contextual similarity into a distance measureXXXXXA