PV-DM

E899024

PV-DM is a neural network-based paragraph vector model that learns distributed representations of sentences and documents by predicting words from both context and document embeddings.

Jump to: Statements Referenced by

Statements (45)

Predicate Object
instanceOf distributed representation model
neural network model
paragraph vector model
unsupervised learning method
abbreviationOf Paragraph Vector – Distributed Memory NERFINISHED
advantage captures word order information in context
provides fixed-length vectors for variable-length texts
application clustering of documents
document classification
information retrieval
sentiment analysis
basedOn neural language model
canUse hierarchical softmax
negative sampling
category document embedding methods
paragraph vectors
comparedWith bag-of-words models
domain natural language processing
representation learning
fullName Paragraph Vector – Distributed Memory NERFINISHED
inputUnit context window of words
document id
inspiredBy distributed memory model of word2vec
introducedBy Quoc V. Le NERFINISHED
Tomas Mikolov NERFINISHED
introducedInPaper Distributed Representations of Sentences and Documents NERFINISHED
languageAgnostic true
learningParadigm unsupervised
learns distributed representations of documents
distributed representations of paragraphs
distributed representations of sentences
optimizationMethod stochastic gradient descent
outputUnit target word
publicationYear 2014
publishedAtConference ICML 2014 NERFINISHED
relatedTo PV-DBOW
word2vec
representationType dense vector
supports variable-length documents
trainingDataRequirement large unlabeled text corpora
trainingObjective predict words using context and document vectors
uses context word embeddings
document embeddings
neural network classifier
vectorSpace continuous vector space

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.