dandye/ai-runbooks/cluster-documents

Document Clustering Skill

Analyze a repository of documents to group them based on content similarity, topic, or purpose. This skill helps organize large collections, identify redundancies, and discover relationships.

Inputs

• PATH - The repository to analyze (e.g., "/repository")
• SIMILARITY_THRESHOLD - (Optional) Float (0.0-1.0), threshold for grouping (default: 0.8)
• VISUALIZATION - (Optional) Boolean, whether to generate a visual representation (default: false)

Workflow

Step 1: Text Processing

Ingest documents from PATH.

• Normalize text (remove stop words, stemming/lemmatization).
• Generate embeddings or TF-IDF vectors for each document.

Step 2: Clustering Analysis

Apply clustering algorithms (e.g., K-Means, DBSCAN) to the document vectors.

• Group documents that meet the SIMILARITY_THRESHOLD.
• Identify outliers or unique documents.

Step 3: Cluster Labeling

Analyze the centroid or representative terms of each cluster to assign a meaningful label (Topic).

Step 4: Output Generation

Generate the clustering report.

• If VISUALIZATION is true, create a scatter plot or dendrogram data.

Required Outputs

A CLUSTERING_REPORT object containing:

• Cluster List: ID, Label, and List of Documents in each cluster.
• Redundancy Report: Sets of highly similar documents (potential duplicates).
• Visualization Data: (If requested) Coordinates for plotting.

Quick Reference

• Purpose: Organize unstructured content and find duplicates.
• Techniques: Text Mining, NLP, Vector Space Models.