Learn Before
A text classification model is designed to categorize sentences into one of three classes: 'Positive', 'Negative', or 'Neutral'. The model works in two stages: first, it generates a unique numerical vector representation for each input sentence. Second, a final component takes this vector and outputs a probability distribution over the three classes. During testing, you observe that for a wide variety of different input sentences, the model consistently outputs probabilities that are very close to uniform (e.g., {'Positive': 0.33, 'Negative': 0.34, 'Neutral': 0.33}). Based on this specific symptom, what is the most direct and likely cause of the problem?
0
1
Tags
Ch.1 Pre-training - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Analysis in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
A text classification model is designed to categorize sentences into one of three classes: 'Positive', 'Negative', or 'Neutral'. The model works in two stages: first, it generates a unique numerical vector representation for each input sentence. Second, a final component takes this vector and outputs a probability distribution over the three classes. During testing, you observe that for a wide variety of different input sentences, the model consistently outputs probabilities that are very close to uniform (e.g., {'Positive': 0.33, 'Negative': 0.34, 'Neutral': 0.33}). Based on this specific symptom, what is the most direct and likely cause of the problem?
Role of the Final Classification Component
A component is added to a model to predict a probability distribution over a set of predefined classes based on an input feature representation. Match each element of this component to its specific function.