Similarity matrix

As explained in Data Without Labels

An affinity matrix is a matrix whose entries quantify pairwise similarities or affinities between data points.

An affinity matrix is a matrix whose entries quantify pairwise similarity between data points. Replacing adjacency-matrix connection values with similarity weights produces values from 0 for complete dissimilarity to 1 for complete similarity, with intermediate values representing degrees of affinity. section 5.3.1

Within graph-based clustering, the matrix connects similarity measurements to the structures used for spectral analysis:

• Represent data points as graph nodes and use stronger edge weights for more similar points. section 5.3 section 5.3.1
• Use the affinity matrix alongside eigenvalues, the degree matrix, and the Laplacian matrix to map data into a lower-dimensional space for clustering. section 5.3
• Interpret affinity-based clustering through connectivity rather than location, making it useful for complex shapes that k-means may separate incorrectly. section 5.3 section 5.9

From graph connections to similarity weights

Start with a graph in which nodes represent data points. An adjacency matrix records whether nodes are connected, whereas an affinity matrix replaces those binary connection values with similarity weights: highly similar points receive larger values, and dissimilar points receive smaller values. A practical exercise is to draw a graph and derive its adjacency and degree matrices before considering the affinity representation. section 5.3.1

Figure 5.4 A graph is a simple representation of data points. The points or nodes are connected by edges if they are very similar (left). The weight is higher if the similarity between data points is high; for dissimilar data points, the weight is less (right).

Use the graph illustration to connect nodes, edges, and edge-weight magnitude with the entries of an affinity matrix. section 5.3.1

How Data Without Labels covers it

How deep it goes

Defines similarity values and links them to spectral clustering, but stops short of detailing matrix construction algorithms.

Examples

Uses weighted graph diagrams, adjacency-matrix figures, complex cluster shapes, and an exercise deriving graph matrices.

What it assumes

Assumes familiarity with data points and matrices while introducing graph nodes, edges, adjacency, degree, and Laplacian matrices.

What you can do after

Build an affinity representation from graph-based similarity weights and relate it to spectral clustering inputs.

How it connects

Connects adjacency matrices, degree matrices, Laplacian matrices, eigenvalues, graph theory, and spectral clustering.

liveBookmanning.com

Read Data Without Labels on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Natural Language Processing in Action, Second EditionA similarity matrix is a square matrix whose entries quantify pairwise similarity between objects.
• 

  liveBookmanning.com
  Graph Neural Networks in ActionA similarity matrix is a matrix whose entries quantify the similarity between pairs of items or representations.

Compare how 3 books explain similarity matrix →

Related concepts

• spectral clusteringused by · Spectral clustering is a clustering method that partitions data using eigenvectors of a graph-based affinity representation rather than only geometric location.
• cosine similarityuses · Cosine similarity measures the directional alignment of two nonzero vectors as the cosine of the angle between them, equivalently their dot product divided by the product of their magnitudes.
• user-item interaction matrixused by · A user-item interaction matrix is a matrix whose rows represent users, columns represent items, and entries encode their interactions, preferences, or ratings.
• adjacency matrixalternative to · An adjacency matrix is a square matrix whose rows and columns represent graph vertices and whose entries indicate edges or encode edge weights between vertex pairs.
• embeddingproduced by · A numerical vector representation of an item or input that encodes features or relationships useful for computational tasks.

This explainer was generated by Manning's knowledge graph from the text of Data Without Labels. The section links point to the parts of the book it draws on.