IMLJD — Indian Matrimonial Litigation Judgment Dataset

A computational dataset of Indian matrimonial litigation judgments for legal NLP, knowledge graph construction, and procedural fairness research.

Dataset: 3,613 cases | Coverage: Supreme Court 2000–2024, Karnataka HC 2018–2024 | Focus: IPC 498A, DV Act, CrPC 482 quashing petitions

---

Key Findings

Finding	Value
SC quash success rate	57.6% (462 of 802 quash petitions)
Karnataka HC quash success rate	39.7% (849 of 2,136 petitions)
Differential (SC vs HC)	+17.9 percentage points
Judicial criticism of misuse (SC)	10.0% of cases
Relatives accused (SC)	7.9% of cases
Settlement at quash stage (SC)	15.2% (122 of 802)

The 17.9-point quash success differential between SC and HC levels suggests systematic pre-filtering: cases reaching the SC are pre-selected for stronger quashing arguments.

---

Dataset Structure

data/
├── parquet/
│   ├── sc_enriched.parquet       # 1,474 SC matrimonial cases 2000-2024
│   ├── hc_29_3.parquet           # 2,136 Karnataka HC 482 petitions 2018-2024
│   ├── sc_enriched.csv           # SC cases (no full text)
│   └── hc_matrimonial.csv        # HC cases
├── extracted/
│   └── sc_<path>.txt             # Full judgment text (SC cases)
└── kg/
    ├── imljd_graph.gexf          # Knowledge graph (Gephi)
    ├── imljd_graph.json          # Knowledge graph (D3/web)
    └── kg_stats.json             # Machine-readable statistics

---

Schema

Supreme Court (sc_enriched.parquet)

Field	Type	Description
case_id	str	Stable identifier
title	str	Case title
petitioner / respondent	str	Party names
decision_date	str	Date of judgment
year	str	Year
disposal_nature	str	Raw disposal (e.g. "Appeal(s) allowed")
case_type	str	quash / appeal / maintenance / bail / other
outcome	str	quashed / allowed / dismissed / settled / disposed / partly_allowed
statutes	str	Pipe-delimited: "IPC 498A | CrPC 482"
mediation_mentioned	bool	Mediation discussed
settlement_mentioned	bool	Settlement / compromise mentioned
omnibus_vague_language	bool	"omnibus/vague allegations" language
relatives_accused	bool	In-laws / extended family named
judicial_criticism_misuse	bool	Court criticises abuse of process
arnesh_kumar_cited	bool	Arnesh Kumar guidelines cited
rajesh_sharma_cited	bool	Rajesh Sharma cited
cited_cases	json	Citation list
word_count	int	Full text word count
allegations_text	str	Extracted facts section
judicial_observations	str	Extracted court observations

Karnataka HC (hc_29_3.parquet)

Field	Type	Description
title	str	Case title (CRL.P/NNNNN/YYYY format)
description	str	Judgment header text
judge	str	Presiding judge
pdf_link	str	eCourts PDF path
decision_date	str	Date
disposal_nature	str	ALLOWED / DISMISSED / DISPOSED / Partly Allowed
outcome	str	Mapped: quashed / dismissed / disposed / partly_allowed
_year	int	Year
_court_name	str	Karnataka High Court

---

Data Sources

Source	URL	Access
SC judgments	s3://indian-supreme-court-judgments/	Public, no auth
HC judgments	s3://indian-high-court-judgments/	Public, no auth

Both are AWS Open Data Registry datasets. No credentials required:

aws s3 ls s3://indian-supreme-court-judgments/ --no-sign-request

---

Reproducing the Dataset

git clone https://github.com/joyboseroy/imljd
cd imljd
pip install -r requirements.txt

# SC metadata (fast)
python3 scripts/03downloadfixed.py --sc --years $(seq 2000 2024)
python3 scripts/enrichsc.py
python3 scripts/fixoutcomes.py

# SC full text (slow — downloads tar archives)
python3 scripts/fetchsclean.py --years 2015 2017 2019 2021 2023

# Karnataka HC 482 petitions
python3 scripts/hcextractv3.py --courts 29_3 --years 2018 2019 2020 2021 2022 2023 2024

# Build knowledge graph
python3 scripts/buildkg.py

---

Requirements

boto3
pyarrow
pandas
pdfplumber
pymupdf
tqdm
rank_bm25
networkx

---

Citation

If you use this dataset, please cite:

@dataset{boseroy2026imljd,
  title     = {IMLJD: Indian Matrimonial Litigation Judgment Dataset},
  author    = {Bose, Joy},
  year      = {2026},
  url       = {https://github.com/joyboseroy/imljd},
  note      = {3,610 cases, Supreme Court 2000-2024 and Karnataka HC 2018-2024}
}

This work extends the legal reasoning framework from:

@article{boseroy2026falkor,
  title  = {FalkorDB-IRAC: Graph-Grounded Legal Reasoning},
  author = {Bose, Joy},
  year   = {2026},
  url    = {https://arxiv.org/abs/2605.14665}
}

---

Ethics

• Public court judgments only — no private communications or FIR data
• Names of parties present as in original public records
• Anonymisation pass recommended before any downstream NLP training
• Framing: procedural fairness research, not case outcome prediction
• Signal flags are descriptive, not diagnostic
• Not suitable for "false case detection" — ground truth does not exist cleanly in this domain

---

Related Work

• ILDC Dataset — 35k SC cases with annotations
• LawSum — 10k+ judgments with summaries
• InLegalNLP — Indian legal NLP benchmark

IMLJD complements these by focusing specifically on matrimonial litigation with outcome labels, signal flags, and a knowledge graph optimised for procedural fairness analysis.