LTprophecyHelp Center
API Reference →

Knowledge Graph

The Knowledge Graph module lets you model causal relationships between business factors, build Bayesian belief networks, and run complex scenario simulations — all without writing code.

Core Concepts

Factor
A named business variable (e.g., Marketing Spend, Sales Headcount, Churn Rate). Factors can be linked to dataset columns for data-backed inference.
Causal Link
A directed relationship between two factors with a configurable strength and lag. Example: Marketing Spend → Pipeline Volume (lag: 30 days, strength: 0.7).
Bayesian Network
A probabilistic graphical model that represents joint probability distributions over factors. LTprophecy learns conditional probability tables from your data.
Scenario Run
A simulation that propagates changes to one or more factors through the causal graph to produce downstream revenue impact estimates.

Building a Causal Graph

  1. Navigate to Knowledge → Factors and click New Factor.
  2. Enter a name, description, and optionally link to a dataset column.
  3. Navigate to Knowledge → Graph to open the visual editor.
  4. Drag factors onto the canvas. Draw arrows between factors to create causal links.
  5. Set the direction (positive/negative), strength (0–1), and lag for each link.
  6. Click Save Graph.

Fitting a Bayesian Network

  1. With your graph saved, navigate to Knowledge → Bayesian Networks.
  2. Click Fit Network and select the causal graph and dataset to use.
  3. Choose discretization bins for continuous variables (default: 5 equal-width bins).
  4. Click Train. Training uses pgmpy and runs on the Celery CPU worker cluster.
  5. Once complete, inspect the learned conditional probability tables under CPT View.

Running a Scenario

  1. Navigate to Knowledge → Scenarios → New Scenario.
  2. Select a fitted Bayesian Network and a scenario type:
    • Factor Drift — shift a factor value by ±X%
    • Volatility Multiplier — expand/contract variance
    • Hard Intervention — set a factor to a fixed value (do-calculus)
  3. Set the drift adjustment or volatility multiplier.
  4. Click Simulate. Results show expected revenue impact with confidence intervals.

Neo4j Integration

For Enterprise plans, the Knowledge Graph is persisted in a Neo4j database alongside PostgreSQL. This enables:

  • Cypher queries for custom graph traversals
  • Multi-hop causal path analysis
  • Graph export to GEXF/GraphML for third-party tools