Optimizing the Performance of Text Classification Models by Improving the Isotropy of the Embeddings Using a Joint Loss Function.
Joseph AttiehAbraham Woubie ZewoudieVladimir VlassovAdrian FlanaganTom BäckströmPublished in: ICDAR (5) (2023)
Keyphrases
- loss function
- classification models
- learning models
- pairwise
- support vector
- learning to rank
- logistic regression
- decision trees
- training data
- feature selection
- risk minimization
- vector space
- empirical risk
- feature set
- reproducing kernel hilbert space
- information retrieval
- convex loss functions
- dimensionality reduction
- models built
- base learners
- attribute selection
- stochastic gradient descent
- boosting framework