如何从R的wordVectors模型提取并导出绘图X/Y坐标至CSV

Got it! Let's walk through how to pull those X/Y coordinates from your wordVectors visualization and save them as a CSV for tools like Tableau. You’ve got two reliable approaches to choose from:

Approach 1: Pull Coordinates Directly from Your Existing Plot #

If you already ran plot() on your trained model to generate the visualization, this is the easiest way—since you’ll get the exact coordinates used in the graph you’re looking at:

# First, save your plot to a variable (replace `your_trained_model` with your actual model name)
my_plot <- plot(your_trained_model)

# Extract the underlying data that powers the plot
plot_coords <- my_plot$data

# Optional: Check the first few rows to confirm it has Word, x, y columns
head(plot_coords)

# Export to CSV—this will save to your current R working directory
write.csv(plot_coords, "word_vector_plot_coords.csv", row.names = FALSE)

The resulting CSV will have all the words paired with their exact X/Y positions from your R plot, ready to import straight into Tableau.

Approach 2: Manually Calculate Dimensionality Reduction (More Flexible) #

If you want to tweak the dimensionality reduction method (like switching between PCA and t-SNE) or adjust parameters, you can compute the coordinates from scratch:

Step 1: Extract the Word Vectors #

First, pull the raw vector matrix from your trained model:

word_vectors <- as.matrix(your_trained_model)  # Replace with your model name

Step 2: Choose Your Dimensionality Reduction Method #

Option A: PCA (Default for wordVectors Plots)

This is what the plot() function uses by default:

# Run PCA to reduce vectors to 2 dimensions
pca_results <- prcomp(word_vectors, scale. = FALSE)

# Format into a data frame with word names and coordinates
pca_coords <- data.frame(
  Word = rownames(word_vectors),
  X = pca_results$x[, 1],
  Y = pca_results$x[, 2]
)

Option B: t-SNE (Better for Clustering Visualization)

If you prefer t-SNE for clearer groupings, you’ll need the Rtsne package:

# Install the package first (only need to do this once)
# install.packages("Rtsne")
library(Rtsne)

# Run t-SNE (adjust perplexity based on your dataset size—10-50 is typical)
tsne_results <- Rtsne(word_vectors, dims = 2, perplexity = 30, max_iter = 1000)

# Format into a data frame
tsne_coords <- data.frame(
  Word = rownames(word_vectors),
  X = tsne_results$Y[, 1],
  Y = tsne_results$Y[, 2]
)

Step 3: Export to CSV #

Save whichever coordinate set you need:

# Export PCA coordinates
write.csv(pca_coords, "word_vector_pca_coords.csv", row.names = FALSE)

# Or export t-SNE coordinates
write.csv(tsne_coords, "word_vector_tsne_coords.csv", row.names = FALSE)

Quick Notes #

• Don’t forget to replace your_trained_model with the actual name of your wordVectors model object (e.g., my_wv_model).
• To find where your CSV saves, run getwd() in R to check your current working directory. You can also specify a full file path in write.csv() (like "C:/MyProjects/word_coords.csv") if you want to save it somewhere specific.

内容的提问来源于stack exchange，提问作者revsequin