Authority Propagation

The core mechanism where a node's authority score is computed based on the quantity and quality of its citing nodes. PageRank variants adapted for law operate on the principle that a citation from a highly authoritative case carries more weight than one from a peripheral decision. The algorithm iterates until scores converge, mathematically expressing the recursive nature of stare decisis—a case is influential if it is cited by other influential cases. This dampens the impact of self-citation loops and citation farms.

Unlike the egalitarian web graph, legal citation networks are strictly hierarchical. Propagation algorithms incorporate jurisdictional filtering to ensure scores reflect binding authority. Citations from higher courts within the same sovereign hierarchy are assigned greater weight than those from persuasive authorities in foreign jurisdictions.

• Vertical propagation: Decisions from appellate courts propagate mandatory authority downward.
• Horizontal attenuation: Citations from coordinate courts or different circuits are weighted lower.
• Sovereign boundaries: Cross-jurisdictional citations are treated as persuasive signals only.

Standard PageRank treats all hyperlinks equally. Legal authority propagation must differentiate between a citation that follows a precedent and one that overrules it. Treatment type classification outputs are used to parameterize edge weights:

• Positive treatment (followed, affirmed): High positive weight, boosting authority.
• Negative treatment (overruled, criticized): Negative or zero weight, diminishing propagated influence.
• Neutral treatment (cited, discussed): Baseline weight for general referential context. This prevents a frequently criticized or overturned case from appearing deceptively authoritative.

Legal authority is not static; it ages and can become obsolete. Temporal citation analysis introduces time-decay factors into the propagation model. Recent citations are weighted more heavily than historical ones to capture the current precedential landscape. This allows the algorithm to detect precedent aging, where a once-seminal case loses influence due to societal or doctrinal shifts, even if it has not been formally overruled. The model can also identify citation cascades triggered by a landmark decision.

Sophisticated propagation occurs on heterogeneous graphs that model more than just case-to-case citations. Nodes represent cases, statutes, constitutions, and administrative regulations. Edges represent distinct relationships: 'interprets,' 'invalidates,' 'amends,' or 'applies.' A statute's authority score is partially derived from the cases that have upheld or struck it down. This multi-entity propagation provides a holistic view of the legal information ecosystem, moving beyond simple case law ranking.

Global PageRank can obscure influential cases within specific doctrinal niches. Community detection algorithms first partition the citation graph into densely connected clusters representing distinct legal topics. Authority propagation is then computed both globally and within each community. This surfaces seminal cases that define a narrow field but may lack broad general influence. A leading case on maritime salvage law can be identified as a community authority even if it is rarely cited by constitutional courts.

A quantitative metric computed over a citation graph that estimates the precedential weight or influence of a legal case. It is typically calculated using iterative algorithms like PageRank variants, which distribute influence based on the quantity and quality of citing sources. A high authority score indicates that a case is frequently cited by other highly authoritative decisions, making it a cornerstone of its doctrinal area. This score is the primary output of authority propagation systems.

A measure of a legal decision's binding or persuasive force, determined by factors including the issuing court's hierarchy level, jurisdictional relevance, and subsequent judicial treatment. Authority propagation algorithms must model this weight as a node attribute to ensure that decisions from a supreme court propagate more influence than those from a trial court. Computational systems often encode this as a prior probability or a damping factor in iterative graph algorithms.

A deep learning architecture designed to operate directly on graph-structured data. In legal AI, GNNs are used to learn node embeddings that capture both a case's intrinsic textual features and its structural role within the citation neighborhood. Unlike traditional PageRank, GNNs can incorporate heterogeneous node features—such as full-text summaries or court metadata—to generate context-aware authority scores that go beyond simple link analysis.

An NLP task that automatically categorizes how a citing case legally treats a cited authority. It assigns labels such as 'overruled,' 'distinguished,' 'followed,' or 'criticized' to each citation instance. This classification is critical for weighting edges in an authority graph; a 'followed' citation should propagate positive influence, while an 'overruled' citation should act as a negative signal, effectively draining authority from the target node.

The algorithmic identification of landmark decisions that serve as the origin points for major legal doctrines. These cases are typically characterized by high out-degree centrality and sustained citation velocity in the authority graph. Authority propagation algorithms naturally surface seminal cases as nodes that consistently rank at the top of influence distributions across multiple iterations and time slices, making them critical seeds for doctrinal lineage tracing.