Weight Profiles¶

Weight profiles control how different types of email interactions are weighted in analysis algorithms.

Overview¶

Email messages have different recipient types:

TO: Direct recipients - the primary audience
CC: Carbon copy - informed but not primary
BCC: Blind carbon copy - hidden recipients

Different analysis goals require different weightings of these recipient types.

Built-in Profiles¶

Influence Profile¶

name: influence
to_weight: 1.0
cc_weight: 0.5
bcc_weight: 0.25

Purpose: Identify actors with organizational influence.

Rationale: Direct (TO) communication indicates stronger relationships. Being addressed directly suggests importance, while CC indicates awareness without direct engagement.

Use cases:

Finding key decision-makers
Identifying informal leaders
Understanding reporting relationships

Information Flow Profile¶

name: information_flow
to_weight: 1.0
cc_weight: 1.0
bcc_weight: 1.0

Purpose: Track how information spreads through the organization.

Rationale: All recipients receive the information equally, regardless of addressing type. The spread of information doesn't depend on whether someone was TO'd or CC'd.

Use cases:

Studying information dissemination
Identifying information hubs
Tracking news/announcement propagation

Coordination Profile¶

name: coordination
to_weight: 0.5
cc_weight: 1.0
bcc_weight: 0.75

Purpose: Identify actors who coordinate activities across groups.

Rationale: CC recipients are often stakeholders being kept informed of coordination activities. Direct recipients may be task-oriented while CC'd parties are coordinators.

Use cases:

Finding project coordinators
Identifying cross-functional connectors
Understanding coordination patterns

Using Profiles¶

Command Line¶

# Use influence profile (default)
commgraph analyze centrality --profile=influence

# Use information flow profile
commgraph analyze centrality --profile=information_flow

# Use coordination profile
commgraph analyze centrality --profile=coordination

Configuration File¶

# .commgraph.yaml
analysis:
  profile: influence

How Weights Are Applied¶

When building the communication graph, each interaction's edge weight is multiplied by the profile weight:

edge_weight = base_weight * profile_weight[recipient_type]

For centrality algorithms like PageRank, these weighted edges determine how "influence" flows through the network.

Example¶

Consider a message from Alice to Bob (TO) and Carol (CC):

With influence profile:

Alice → Bob: weight = 1.0 × 1.0 = 1.0
Alice → Carol: weight = 1.0 × 0.5 = 0.5

With information_flow profile:

Alice → Bob: weight = 1.0 × 1.0 = 1.0
Alice → Carol: weight = 1.0 × 1.0 = 1.0

With coordination profile:

Alice → Bob: weight = 1.0 × 0.5 = 0.5
Alice → Carol: weight = 1.0 × 1.0 = 1.0

Choosing a Profile¶

If you want to find...	Use profile
Key influencers and decision-makers	`influence`
Information hubs and broadcasters	`information_flow`
Coordinators and project managers	`coordination`

Impact on Results¶

The choice of profile can significantly affect results:

Example: Executive Assistant¶

An executive assistant who is often CC'd on executive communications:

influence profile: Lower ranking (CC has lower weight)
coordination profile: Higher ranking (CC has higher weight)
information_flow: Moderate ranking (all weights equal)

Example: Project Manager¶

A project manager who CC's many stakeholders:

influence profile: Moderate (their TO recipients score high)
coordination profile: High (their CC recipients score high)
information_flow: High (all communication counts equally)

Advanced: Custom Profiles¶

For programmatic use, you can define custom weight profiles:

import "github.com/grokify/commgraph/analysis"

profile := &analysis.WeightProfile{
    Name:      "custom",
    ToWeight:  1.0,
    CCWeight:  0.75,
    BCCWeight: 0.1,
}

results, err := analysis.Centrality(store, &analysis.CentralityOptions{
    Algorithm: analysis.PageRank,
    Profile:   profile,
})

Best Practices¶

Start with influence: The default profile works well for most organizational analysis.
Compare profiles: Run analysis with different profiles to see how results change.
Consider context: Choose the profile that matches your analysis question.
Document your choice: When sharing results, note which profile was used.