This dataset is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 licence (CC BY-NC-SA 4.0): research, academic, personal and other non-commercial use is permitted with attribution; adaptations must be shared under the same licence; commercial use is not permitted. The original public source data remains under the Government Open Data License - India (GODL-India). Commercial use: contact surya@judicialmind.ai or hello@judicialmind.ai.

KCC-Pairs

Licence: CC BY-NC-SA 4.0 — research and non-commercial use only; derivatives must be shared under the same licence. Commercial use: contact surya@judicialmind.ai or hello@judicialmind.ai

Access is approved automatically when the email you enter matches your Hugging Face account email and belongs to the organisation you name; other requests are reviewed manually.

Farmer question -> call-centre advisor answer pairs from India's Kisan Call Centre (KCC) logs: 4,996,402 question-answer pairs and 502,060 question-question pairs, cleaned, de-duplicated and scrubbed of personal data, with year- and state-held-out splits, plus a fixed retrieval benchmark (5,000 test queries against 100,000 KCC answers, most of them cross-lingual: English question, answer in Hindi, Marathi, Gujarati, Punjabi or Kannada) and an evaluation script.

Headline result: KCC-Embed-0.6B, trained on these pairs, reaches R@10 0.417 on the retrieval benchmark, about twice the best baseline (pplx-embed-v1-0.6b 0.207; BM25 0.122).

Research data, not advice. Answers were given by Kisan Call Centre advisors and may contain pesticide doses or product names that are outdated, region-specific or wrong. Do not use them as agronomic advice.

Quickstart

The repository is gated: accept the terms on the dataset page, then huggingface-cli login.

from datasets import load_dataset

# question -> answer pairs (one JSON object per line); pick the files you need
test = load_dataset("judicialmind/KCC-Pairs", data_files={"test": "data/test.jsonl"}, split="test")
print(test.num_rows)                      # 546237
row = test[0]
print(row["query"], "->", row["answer"], f"({row['query_lang']} -> {row['answer_lang']}, {row['state']})")

# retrieval benchmark: queries with gold answer ids, and the answer corpus
q = load_dataset("judicialmind/KCC-Pairs", "retrieval_queries", split="test")    # 5,000 queries
corpus = load_dataset("judicialmind/KCC-Pairs", "retrieval_corpus", split="test")  # 100,000 answers
print(q[0]["query"], q[0]["positives"])

With pandas or DuckDB on a local copy (huggingface-cli download judicialmind/KCC-Pairs --repo-type dataset --local-dir KCC-Pairs):

import duckdb, pandas as pd

dev = pd.read_json("KCC-Pairs/data/dev.jsonl", lines=True)
print(dev.groupby("answer_lang").size().sort_values(ascending=False).head())

print(duckdb.sql("""
    SELECT state, count(*) AS pairs
    FROM read_json_auto('KCC-Pairs/data/test.jsonl')
    GROUP BY state ORDER BY pairs DESC
""").df())

Evaluate a retrieval model on the benchmark (prints JSON with overall and per-language metrics):

pip install -r requirements.txt
python evaluate.py --model bm25                               # lexical baseline
python evaluate.py --model judicialmind/KCC-Embed-0.6B        # dense model, mean pooling + cosine
python evaluate.py --model BAAI/bge-m3 --sentence-transformers
python evaluate.py --run my_run.jsonl                          # score your own ranking

A run file has one line per query, {"id": "kcc-q-...", "ranked": ["kcc-ans-...", ...]} (best first, up to 100 ids); --save-run FILE writes one from any model. --split dev uses the dev queries and corpus.

Files

file	rows	contents
`data/train.jsonl`	3,954,756	years before 2024, states not held out
`data/dev.jsonl`	41,736	1% of train query clusters
`data/test.jsonl`	546,237	years 2024-2025 or held-out states (Madhya Pradesh, Himachal Pradesh); query clusters unseen in train
`data/test_seen.jsonl`	453,673	test-period pairs whose query cluster also occurs in train
`data/train_qq.jsonl`	502,060	query -> query positives (same answer and crop)
`retrieval/test_queries.jsonl`	5,000	retrieval test queries, drawn from `test`
`retrieval/test_corpus.jsonl`	100,000	answer corpus for the test queries
`retrieval/dev_queries.jsonl`	2,000	retrieval dev queries, drawn from `dev`
`retrieval/dev_corpus.jsonl`	100,000	answer corpus for the dev queries
`evaluate.py`		retrieval metrics (BM25, transformers, sentence-transformers or a run file)

Calls dropped before pairing: weather 9,757,908, short_answer 1,046,955, market_price 994,081, helpline_only 320,641, pmkisan_status 236,669, empty 77,574, call_noise 61,157, non_answer 11,341.

Schema

Pairs (data/train|dev|test|test_seen.jsonl):

field	type	description
`id`	string	pair id (`kcc-pair-...`)
`query`	string	farmer question as typed by the call-centre advisor (mostly short English summaries)
`answer`	string	advisor answer (often in the farmer's language)
`query_lang`, `answer_lang`	string	ISO 639-1 code (`hinglish` = Hindi in Latin script; empty when undetermined)
`state`, `district`	string	state and district of the call, as in the source (upper case)
`district_lgd`	int	LGD district code (null in this release)
`crop`, `category`	string	crop and crop category from the call record (a few rows carry the source's numeric code)
`query_type`	string	KCC query type, e.g. `Plant Protection`, `Water Management`
`season`	string	`KHARIF`, `RABI`, `JAYAD` or null
`year`, `month`	int	call date
`split`	string	`train`, `dev`, `test` or `test_seen`
`scrubbed`	bool	personal-data scrub applied
`masked`	bool	the answer carried a mask (`[PHONE]`, `[EMAIL]`, `[ID]`, `[NAME]`)
`dup_count`	int	number of identical calls collapsed into this pair
`qkey`	string	normalised query text (query cluster key)
`kind`	string	`qa`
`hard_negative`	string	train only: a different answer to a similar question, for contrastive training

Query pairs (data/train_qq.jsonl): id, query, positive (a different question with the same answer and crop), crop, kind (qq), split.

Retrieval queries (retrieval/*_queries.jsonl): id, query, lang, positives (list of corpus ids; 1.55 per test query on average), answer_lang, cross_lingual, crop, state, split; also hard_negatives, sibling_negatives, template_id (empty here), ministry, generator (human: real KCC questions), checked, check. Retrieval corpus (retrieval/*_corpus.jsonl): id, text, crop.

Test query language pairs: en->hi 2,022, en->en 1,797, en->mr 700, en->gu 292, en->pa 79, en->kn 73, other 37.

Results

Retrieval test set (5,000 queries with unseen query clusters, years 2024-25 and held-out states; 100,000-answer corpus), scored with evaluate.py (python evaluate.py --model bm25 reproduces the bm25 row):

model	R@1	R@10	nDCG@10
bm25	0.078	0.122	0.093
perplexity-ai/pplx-embed-v1-0.6b	0.115	0.207	0.148
Qwen/Qwen3-Embedding-0.6B	0.097	0.171	0.123
BAAI/bge-m3	0.097	0.169	0.122
judicialmind/KCC-Embed-0.6B	0.206	0.417	0.287

Coverage

10 best-covered states: Rajasthan, Maharashtra, Gujarat, Bihar, Punjab, Odisha, Karnataka, Haryana, Madhya Pradesh, Himachal Pradesh; calls from 2006 to 2025.
The source holds 27,754,152 calls from the public data.gov.in snapshot.
Phone numbers, e-mails, ID numbers and personal names were masked ([PHONE], [EMAIL], [ID], [NAME]); 852619 distinct answers carried a mask.

Source data

Derived from 1 resources published on the Open Government Data Platform India (data.gov.in), snapshot May 2026, under the Government Open Data License - India. GODL-India does not cover personal information; personal identifiers were masked before release.

Providers: Ministry of Agriculture and Farmers Welfare.

Attribution: Department of Agriculture and Farmers Welfare (Ministry of Agriculture and Farmers Welfare). "Kisan Call Centre (KCC) farmer queries". data.gov.in resource cef25fe2-9231-4128-8aec-2c948fedd43f, https://api.data.gov.in/resource/cef25fe2-9231-4128-8aec-2c948fedd43f. Licensed under the Government Open Data License - India (https://data.gov.in/government-open-data-license-india). Snapshot: May 2026.

Original data: published by Department of Agriculture and Farmers Welfare, Ministry of Agriculture and Farmers Welfare (Kisan Call Centre) on the Open Government Data Platform India (data.gov.in) and, where marked, NITI Aayog's National Data and Analytics Platform (NDAP), under the Government Open Data License - India (GODL-India). The original public data remains under GODL-India; the CC BY-NC-SA 4.0 licence of this repository covers JudicialMind's compilation, derived fields, annotations, benchmarks and models.

Citation

@misc{judicialmind2026kccpairs,
  title={KCC-Pairs: Kisan Call Centre question-answer pairs and cross-lingual retrieval benchmark},
  author={JudicialMind},
  year={2026},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/judicialmind/KCC-Pairs}}
}

License

CC BY-NC-SA 4.0 (see LICENSE): research, academic, personal and other non-commercial use, with attribution to JudicialMind and to the original data publishers; adaptations must be shared under the same licence. Commercial use requires a separate licence.

Commercial use: contact surya@judicialmind.ai or hello@judicialmind.ai.

Downloads last month: 18

Total file size:

3.67 GB

Collection including judicialmind/KCC-Pairs

India Open Government Data — Bharat Agriculture Datasets

Collection

Research and non-commercial use (CC BY-NC-SA 4.0). Commercial: surya@judicialmind.ai / hello@judicialmind.ai • 3 items • Updated 1 day ago