在R语言K-means聚类图添加标签及解读聚类的技术问询

Hey there! Let's tackle your two K-means clustering questions one by one—super common pain points when working with text clustering, so I get where you're coming from.

Your original text() call was pointing to the wrong data source (you used m_norm instead of the PCA projection coordinates). Here's a corrected, cleaner approach to plot your PCA results and add document labels clearly:

# First, compute PCA and store the result for easy reuse
pca_result <- prcomp(m_norm)

# Plot the PCA points with cluster colors (pch=16 makes points more visible)
plot(
  pca_result$x, 
  col = cl$cluster, 
  pch = 16, 
  main = "K-Means Clustering (5 Clusters) - PCA Projection",
  xlab = "Principal Component 1",
  ylab = "Principal Component 2"
)

# Add document labels (row names) to each data point
text(
  x = pca_result$x[,1], 
  y = pca_result$x[,2], 
  labels = rownames(m_norm), 
  cex = 0.7,  # Shrink label size to avoid overlap
  pos = 3     # Place labels above points
)

If your document names are too long, you can shorten them with abbreviate(rownames(m_norm), minlength=5) in the labels argument to keep the plot readable. For even better interactivity (hover to see labels), try the plotly package:

library(plotly)
plot_ly(
  x = pca_result$x[,1], 
  y = pca_result$x[,2], 
  color = factor(cl$cluster), 
  text = rownames(m_norm), 
  type = "scatter", 
  mode = "markers"
)

Let's break this down into two parts:

What does table(cl$cluster) mean? #

The output numbers are simply the count of documents in each cluster. For example, if you see:

1   2   3   4   5 
 85  92  78  80  65 

That means Cluster 1 has 85 documents, Cluster 2 has 92, and so on. The total should add up to your full dataset size (400+ documents).

How to interpret the actual content of each cluster? #

To understand what each cluster is about, you need to look at the most representative terms (words) for each group. Here are two reliable methods:

Method 1: Extract top-weighted terms per cluster #

K-means gives you cluster centers (cl$centers) which represent the average TF-IDF weight of each term in the cluster. We can sort these weights to find the most important words:

# Get the list of terms from your document-term matrix
terms <- colnames(dtm_tfxidf)

# Extract top 10 terms for each cluster
top_terms <- lapply(1:5, function(cluster_num) {
  # Get center weights for the cluster
  cluster_weights <- cl$centers[cluster_num, ]
  # Sort terms by weight (highest first)
  sorted_terms <- sort(cluster_weights, decreasing = TRUE)
  # Grab top 10
  head(sorted_terms, 10)
})

# Name the list for clarity
names(top_terms) <- paste("Cluster", 1:5)

# Print the results
top_terms

If Cluster 1's top terms are "climate", "emission", "policy", you can infer that cluster focuses on climate change policy documents.

Method 2: Inspect actual documents in a cluster #

To validate your term-based conclusions, you can directly view which documents belong to a cluster. Assuming your original documents are stored in a data frame/vector called docs_data:

# Add cluster labels to your document dataset
docs_data$cluster <- cl$cluster

# View all documents in Cluster 1, for example
subset(docs_data, cluster == 1)

This lets you read sample documents and confirm the cluster's theme matches the top terms.

内容的提问来源于stack exchange，提问作者daveyjones