Transfer Learning Approach

You can also work with a transfer learning approach for the word embeddings by adding a pretrained embedding matrix. We will use the previous model of the English wikipedia. Note that the package requires the dimensions of word embeddings and docs to be equal<

# Use a pretrained embedding matrix
model <- read.word2vec(file = 'model.bin', normalize = TRUE)

# Build a PV-DM model with transfer learning
pvdm <- paragraph2vec(x = doc,
                      type = "PV-DM",
                      dim = 300,
                      iter = 20,
                      min_count = 5,
                      lr = 0.05,
                      threads = 1,
                      embeddings = as.matrix(model))

Document Embeddings for Predictive Modeling

Get the document embeddings that can be used in a predictive model.

# Retrieve document embeddings for predictive modeling
embeddings <- as.matrix(pvdm, which = "docs")