Legal Intelligence: Case Law Connections

The Law is a Graph. Every new case cites an old case. Every statute refers to another statute. Every judge has a history of specific rulings. A lawyer doesn't just need to "Find a Document"; they need to Find a Path of Logic from a current situation back to a binding precedent. Graph RAG is built for this "Citation Reasoning."

In this lesson, we will look at how to build a Legal Knowledge Graph. We will learn how to extract [:CITES] and [:OVERRULES] relationships from legal text and how to use an AI agent to answer: "Is there any case where this specific argument was used and was successfully upheld on appeal?"

---

1. The Legal Graph Schema

• (:Case) {date, jurisdiction, outcome}
• (:Statute) {section_id, language}
• (:Judge) {biography, leanings}
• (:Legal_Concept) (e.g., "Habeas Corpus", "Negligence")

---

2. The Power of "Citation Strength"

In a Graph RAG system, not all citations are equal.

• Direct Citation: Case A refers to Case B.
• Influence Score: If Case B is cited by 1,000 other cases, its PageRank (Module 11) is high. It is a "Landmark Case."

AI Reasoning: When a lawyer asks a question, the AI prioritizes landmark cases over obscure ones, ensuring that the legal advice is grounded in the most authoritative precedents.

---

3. Detecting "Overruled" Paths (Conflict Detection)

One of the most dangerous mistakes a lawyer can make is citing a case that has been Overruled. In a graph, this is a simple Path Trap.

• Edge: (Case_A)-[:OVERRULES]->(Case_B).
• If the AI retrieves Case B for an answer, it must check for any incoming [:OVERRULES] edges. If it finds one, it must alert the user: "Warning: This case was overruled by Case A in 2023."

graph TD
    C1[Case: Landmark 1990] -->|Cited By| C2[Case: Mid 2010]
    C3[Case: Recent 2024] -->|OVERRULES| C2
    C2 -->|Supports| Arg[Legal Argument]
    
    style C3 fill:#f44336,color:#fff
    note[The AI must follow the 'Overrules' link to invalidate the argument]

---

4. Implementation: Finding Precedent Paths in Cypher

MATCH (current:Case {id: $id})-[:CITES*1..3]->(precedent)
MATCH (precedent)-[:RELIED_ON_BY]->(outcome)
WHERE precedent.jurisdiction = 'California'
RETURN precedent.title, outcome.summary;

// This query finds the 3-hop 'Ancestry' of a current case 
// to see which California precedents it is built upon.

---

5. Summary and Exercises

Legal Graph RAG is about Logical Lineage.

• Citation Traversal is the primary method of legal research.
• Landmark Detection (PageRank) filters out irrelevant noise.
• Conflict Detection (:OVERRULES) prevents quoting invalid law.
• Concept linking allows for "Analogous Reasoning" across different domains.

Exercises

1. Schema Task: If Case A is a "Supreme Court" case and Case B is a "District Court" case, how do you mark the "Authority" of the relationship between them?
2. The "Hidden" Precedent: How would you find a case that doesn't cite your target directly but uses the same 5 "Legal Concepts"? (Hint: Search for nodes sharing 5 shared concept neighbors).
3. Visualization: Draw a "Case" node. Draw 3 arrows pointing out to older cases. Draw 1 "Red" arrow pointing inward from a newer case.

In the next lesson, we will look at vertical data: Biomedical Research: Tracking Disease Paths.