AI-ordlista
Den kompletta ordlistan över AI
Latent Dirichlet Allocation (LDA)
Generative probabilistic model that discovers abstract topics in a collection of documents by assuming that each document is a mixture of topics and each topic is a distribution of words.
Dirichlet Distribution
Multivariate probability distribution over the simplex used as a prior distribution in mixture models like LDA to model topic proportions.
Topic Coherence
Qualitative measure evaluating the semantic cohesion of generated topics by analyzing the co-occurrences of the most probable words in a reference corpus.
Document-Topic Matrix
Matrix θ where each row represents a document and each column the probability distribution of topics in that document, main result of LDA.
Word-Topic Matrix
Matrix φ representing the probability distribution of words for each topic, indicating the relevance of each word to the different discovered themes.
Hyperparameters Alpha Beta
Parameters of the Dirichlet distribution where α controls the dispersion of topics in documents and β the dispersion of words in topics.
Collapsed Gibbs Sampling
Optimized variant of Gibbs Sampling where parameters θ and φ are analytically integrated, significantly accelerating LDA model convergence.
Online LDA
Stochastic version of LDA processing documents in mini-batches to enable application on massive corpora with reduced memory complexity.
Hierarchical Dirichlet Process
Non-parametric extension of LDA that automatically infers the optimal number of topics from the data without specifying this value a priori.
Convergence Monitoring
Process of tracking the stability of LDA model parameters between successive iterations to determine when the algorithm has reached a stationary state.