1 Introduction
2 Related work
2.1 Learning at entity level
2.2 Error-weighted predictions and clustering entities
2.3 Reducing memory footprint with lossy counting
3 Methods
3.1 Entity-centric learning
3.1.1 Entity-centric modeling of the stream
3.1.2 An ensemble with two voting members
3.1.3 The entity-centric ensemble member
3.1.4 The entity-ignorant ensemble member
3.1.5 Ensemble variants based on weighting
3.2 Memory reduction
3.2.1 Entity management with lossy counting
3.2.2 Replacing entity-centric models with text-ignorant models
4 Experiments
Name | #Ent. | #Inst. | #Feat. | #Classes |
---|---|---|---|---|
tools | 33,990 | 1,417,499 | 10,000 | 5 |
watches | 78,220 | 487,907 | 10,000 | 5 |
bars5 | 25,110 | 2,224,710 | 10,000 | 5 |
barsFull | 59,372 | 4,198,061 | 10,000 | 5 |