unsupervised learning and evaluation

unsupervised machine learning

concept involves teaching computers to discover patterns and relationships in language without explicit guidance or labeled examples.

topic modeling

assigns to a corpus based on the words present, essential for categorizing documents in a data-rich world.

deep learning

subset of machine learning mimicking human brain workings through artificial neural networks

transformers

introduced in 2017 for multilingual translation, allows parallel processing, encodes word positions, and simplifies lookup.

BERT (bidirectional encoder representations from transformers)

pre-trained contextual model setting benchmarks in various nlp tasks; reads text in both directions, facilitates transfer learning, and is versatile for tasks like text classification and sentiment analysis

evaluation

crucial for ensuring model performance, avoiding misinterpretations, and maintaining transparency.

accuracy

ratio of correctly predicted instances to total instances

recall

ratio of true positives to true positives and false negatives

precision

ratio of true positives to true positives and false negatives

confusion matrix

illustrates correct and incorrect predictions

ROC curve

plots true positive rate against false positive rate

class imbalance

high accuracy but low f1-score

overfitting

high training accuracy but low validation accuracy

underfitting

low training and validation accuracy