{
  "field_id": "F034",
  "field": "cluster analysis",
  "directions": [
    {
      "direction_id": "G0",
      "name": "Representation and relational structure for cluster recovery",
      "scope": "Learn, construct, or compress features and relationships between observations so that latent clusters can be distinguished and recovered; fully encompasses joint representation-and-assignment learning, self-expressive subspace relationships, and cluster recovery after dimensionality reduction."
    },
    {
      "direction_id": "G1",
      "name": "Integration of complementary clustering evidence",
      "scope": "Reconcile complementary evidence from different views or existing clustering results into a shared cluster structure; fully encompasses jointly available views, recovery when views are missing, and consensus aggregation of base partitions."
    },
    {
      "direction_id": "G2",
      "name": "Clustering from graph relationships",
      "scope": "Use graph connections, attributes, or relational objectives to determine node partitions; fully encompasses spectral and graph operators, learned node clustering, and discrete graph-partition optimization."
    },
    {
      "direction_id": "G3",
      "name": "Neighborhood-defined cluster discovery",
      "scope": "Form and separate clusters according to local concentration, connectivity, and neighborhood geometry, including irregularly shaped groups."
    },
    {
      "direction_id": "G4",
      "name": "Representative-based partition fitting",
      "scope": "Form clusters by fitting probabilistic components or selecting centers, prototypes, and medoids, and study fitting quality and computation."
    },
    {
      "direction_id": "G5",
      "name": "Clustering under outlier contamination",
      "scope": "Recover useful partitions or components in the presence of outlying or adversarial observations, without allowing anomalous observations to dominate estimation and assignments."
    },
    {
      "direction_id": "G6",
      "name": "Fairness-constrained clustering",
      "scope": "Make cluster memberships, centers, costs, or hierarchies satisfy explicit group-fairness requirements."
    },
    {
      "direction_id": "G7",
      "name": "Clustering guided by supplied side information",
      "scope": "Use limited supplied information about memberships or relationships between observations to guide cluster assignments and recovery."
    },
    {
      "direction_id": "G8",
      "name": "Temporal-pattern clustering",
      "scope": "Define groups according to the alignment, shape, dynamics, or predictive behavior of time series or subsequences."
    },
    {
      "direction_id": "G9",
      "name": "Structured and intelligible cluster descriptions",
      "scope": "Make clustering results more than uncharacterized flat assignments by presenting nested relationships or intelligible group definitions and explanations."
    },
    {
      "direction_id": "G10",
      "name": "Assessment and validation of cluster structure",
      "scope": "Determine whether data support clusters and how many, or measure and compare completed clusterings and assess the trustworthiness of their evaluation basis."
    },
    {
      "direction_id": "G11",
      "name": "Clustering with restricted data access and disclosure",
      "scope": "Obtain useful clustering when raw observations cannot be freely pooled or released."
    }
  ]
}